Wright 93

The Dynamic Simulation and Control of Aluminium
Smelting Cells
by
A. Wright B.Sc.
Department of Chemical and Process Engineering

University of Newcastle-upon-Tyne
Newcastle-Upon-Tyne, NE1 7RU
England
Submitted in partial fulfilment of the requirements for the degree of Doctor

of Philosophy at the University of Newcastle-Upon-Tyne
NEWCASTLE UNIVERSITY LIBRARY

093 51505 4
June, 1993
ABSTRACT
The Hall-Heroult process for the electrolytic production of aluminium from

alumina is a costly and difficult to control process that has remained little
changed since the early 1900's. A decreasing trend in the real profitability
of the process since 1930 has made it necessary for aluminium smelting
companies to reduce the expenditure in all aspects of the process in order
to remain viable. The most significant proportion of the costs of production
is the utilisation of the electrical energy required to produce the aluminium
but improvements such as the rebuilding of the cell superstructure in
modern low resistance materials are precluded by the capital costs
involved. It is generally only possible to make improvements by changes
in operational procedure and the control strategy.
The introduction of computer control and data logging systems has

reduced the manual involvement in cell operation and has allowed control
strategies to be standardised on all cells within a smelter. Although the
data logging facilities have increased the amount of data that can be
collected and improved the understanding of the operation of aluminium
smelting cells, the control of remains difficult due to the lack of data that
can be continuously monitored, in particular the alumina concentration in
the electrolyte.
In this work, a mathematical model for the dynamic simulation of

aluminium smelting cells is developed. A simulation program is then
written incorporating the exact control algorithms from Anglesey
Aluminium's GELTROl computer control system for half-break cells. The
aim of the simulation is to study the effects of different operating
conditions and control strategies upon the operation of the cell. The
simulation is developed to be modular in nature allowing different control
systems and cell models to be easily incorporated and tested.
The model is evaluated against data from Anglesey Aluminium's half-

break cells and is successful in predicting the behaviour of these cells.
The simulation is also used in plant trials to investigate the effects of
different metal tapping schedules.
Acknowledgements
I would like to express my sincere thanks to the following people who have
enabled me to carry out this research:
My supervisor, Dr Allen Wright for his continual support, encouragement

and advice during the course of this research.
The staff of the Department of Chemical and Process Engineering for their
help and advice.
George Guelfo, David Salt, Tony Sargent and Mike Wilson of Anglesey
Aluminium for their support and practical assistance in developing the
model and their patience when showing me the ropes of aluminium
production.
The technical staff of Anglesey Aluminium also for their advice and
assistance.
The UK Science and Engineering Research Council and Anglesey

Aluminium for financial support.
Finally, I would like to thank my wife, Suzanne for her moral support and
encouragement during the overly long course of this research.
To my wife, Suzanne and our son, Toseland ...
Contents
Page
1.0 History of the Hall-Heroult Process 1
1.1 Main Costs Of Production 1
1.1.1 Fundamental relationships 2
1.2 Types of Cell 3
1.3 Operation of a Prebake Cell 6
1.3.1 Data Sampling 6
1.3.2 Alumina feed 7
1.3.3 Voltage control 7
1.3.4 Anode effect termination 8
1.3.5 Metal tapping 9
1.3.6 Addition of raw materials other than alumina 10
1.3.7 Anode change 11
1.3.8 Sampling 12
2.0 Process Modelling 15
2.1 Model Development 16
2.2 Types of Model 18

2.2.1 Lumped parameter models 18
2.2.2 Distributed parameter models 18
2.2.3 Static and dynamic models 19
2.3 Choice of Model 20
3.0 Mathematical Models of Aluminium Reduction Cells 21
3.1 Static Models 21

3.1.1 Model of M.C. Richard 21
3.1.2 Model of B. Sulmont and G. Hudault 21
3.2 Dynamic Models 22

3.2.1 Model of S. R Blake 22
3.2.2 Model of A. Ek and G. E. Fladmark 22
3.2.3 Model of T. Hashimoto and H. Ikeuchi 23
3.3 Models for the Dissolution of Alumina 23

3.3.1 Zeroth order models 24
Page
3.3.2 First order models 25
3.3.3 Heat transfer models 25
3.3.4 Addition of alumina to electrolyte. 27
3.4 Models for the Evolution of Fluorine from Reduction
Cells 28
3.4.1 Model of W.E. Haupin 29
3.4.2 Gaseous fluoride 29
3.4.3 Particulate fluoride by volatilisation 31
3.4.4 Particulate fluoride by entrainment 32
3.5 Models for the Loss of Current Efficiency 33
3.5.1 Empirical relationships 37
3.5.2 Model of Robl et al 38
3.5.3 Model of Evans et al 41
3.5.4 Model of B. Lillebuen et al 42
3.5.5 Calculation of mass transfer coefficients 43
3.5.6 Model of Vasiliadis 46
3.5.7 Model of Haupin 49
3.6 Models for the Estimation of Frozen Bath Shape 51
3.6.1 Determination of heat transfer coefficients 54
3.6.2 Model of Haupin 54
3.6.3 Model of A. Solheim and J. Thonstad 55
3.6.4 Model of B.J. Welch 56
3.7 Models for the Voltage Drop across the Cell 58
3.7.1 Electrical conductivity 61
3.7.2 Reversible decomposition potential and
electrode overpotential 64
3.7.3 Cathodic overpotential 66
4.0 Development of a Dynamic Model 75

4.1 Reactor Modelling 75
4.2 Process Inputs 76
4.2.1 The breaker action 76
4.2.2 Anode beam position 77
4.2.3 Line current 77
4.2.4 Anode change 77
II
Page
4.2.5 Stochastic disturbances 77
4.3 Alumina Dissolution 78
4.4 Electrolyte Mass Balance 81
4.4.1 Bath volatilisation 82
4.4.2 Gaseous emissions 82
4.4.3 Particulate emissions by volatilisation 84
4.4.4 Aluminium dissolution and back reaction 88
4.4.5 Current efficiency during anode effect 91
4.4.6 Current efficiency losses at the cathode 93
4.4.7 Overall cu rrent efficiency 94
4.4.8 Mass balances 95
4.4.9 Material in suspension and as sludge 95
4.4.10 A13+ ions in solution 97
4.4.11 0 2 - ions in solution 98
4.4.12 Aluminium metal in electrolyte 99
4.4.13 Aluminium in the metal pad 99
4.4.14 Sodium fluoride 100
4.4.15 Calcium fluoride 101
4.4.16 Alumina in solution 102
4.4.17 Aluminium fluoride 102
4.4.18 Overall mass balance on electrolyte 103
4.5 Electrolyte Conductivity 103
4.6 Electrolysis Charge Balance: Reaction Modelling 105
4.6.1 Determination of Tafel constants and mass
transfer coefficient 111
4.6.2 Reactor modelling 114
4.6.3 Anode effect 117
4.7 Cell Voltage, Energy Consumption and Cell
Resistance 120
4.8 Heat Balance 121
4.9 Electrolysis Heat Balance 122
III
Page
4.10 Heat Losses 124

4.10.1 Heat transferred to the frozen sidewalls 124
4.10.2 Heat transferred from metal pad to cathode
block 125
4.10.3 Heat transferred from electrolyte to cavity 126
4.10.4 Heat transferred from electrolyte to anodes 129
4.11 Pot Heat Balance 132

4.11.1 Heat of dissolution 134
4.11.2 Specific heat of feed materials 135
4.12 Frozen Sidewalls 136

4.12.1 Heat transfer coefficients 139
4.13 Anode Change 139

4.13.1 Heat losses 140
4.13.2 Resistance of anode block 141
4.14 Phyico-Chemical Relationships 143

4.14.1 Alumina solubility 143
4.14.2 Aluminium solubility 144
4.14.3 Density 144
4.14.4 Liquidus temperature 144
4.14.5 Viscosity of the electrolyte 145
4.15 Summary 145
5.0 The Control System 149
5.1 Alumina Concentration Control 150
5.2 Removal of Aluminium Product 150
5.3 Electrolyte Composition 151
5.4 Anode to Cathode Distance 151
5.5 Current Distribution 152
5.6 Voltage and Temperature Control 152

5.6.1 Frozen sidewalls 153
5.7 Manual Control 153
5.8 Manual Control of Alumina Concentration 154
IV
Page
5.9 Manual Control of Voltage 154

5.10 Manual Control Strategy 155
5.11 Automatic Control 155
5.12 Distributed Control 156
5.13 The CElTROl System 156
5.13.1 The CElTROl unit: a microprocessor 157
5.13.2 The communications buffer 157
5.13.3 The supervisory computer 158
5.13.4 The development/backup computer 159
5.14 Automatic Control Strategies 159
5.15 Half Break Control Strategy 160
5.15.1 Automatic voltage control, AVC 160
5.15.2 Normal operation 161
5.15.3 Noisy cells 161
5.15.4 Anode effect 162
5.15.5 Metal tap 162

5.16. 1 Normal operation 163
5.16.2 Search routine 163
5.16.3 Starve routine 164
5.17 Anode Effect Termination 165
5.18 Point Feeder Control Strategies 165

5.19 Anode Effect Termination Routine 166
5.20 Alternative Strategies 166
5.22 Open loop Alumina Feed Strategies 167

5.22.1 low alumina concentration 167
5.22.2 High alumina concentration 167
5.23 Closed loop Alumina Feed Strategies 168

5.23.1 Half break cells: adaptive feed strategy 168
v
Page
5.23.2 Point feed cells: adaptive feed strategy 168
5.23.4 Overfeed/Underfeed strategy (O/U strategy) 169
5.24 ACD Control Strategies 171
5.24.1 Fixed ACD 171
5.24.2 ACD optimisation 171
5.25 Summary 172
6.0 Dynamic Simulation of an Aluminium Reduction Cell 173

6.1 Choice of Programming language 173
6.2 Initialisation 174
6.3 Presentation of Results 175
6.4 Numerical Solution of Differential Equations 176
6.5 Simulation of Current Control System (CElTROl) 176
6.6 Validation of the Simulation 177

6.6.1 Current efficiency 178
6.6.2 Energy efficiency 179
6.6.3 Cell temperature 180
6.6.4 Weight Ratio NaF/AIF3 180
6.6.5 Anode to Cathode Distance (ACD) 180
6.6.6 Parameter trends 182
6.6.7 Alumina concentration 182
6.6.8 Cell voltage 183
6.6.9 Bath depth 184
6.6.10 Weight ratio NaF/AIF3 (Ratio) 185
6.6.11 Temperature 186
6.7 Plant Trials 190
6.8 Conclusions 191
7.0 Recommendations For Further Work 193
7.1 Inferential Control 193
7.2 Development of the Real Time Simulation (RTS) 193
VI
Page
7.3 Tuning the Model for Control 198

7.4 Alumina Concentration Meter 199
Appendices 203
A Initialisation of the Simulation Program 203

A.1 Initialisation Files 203
A.2 The User Interface for the Simulation Program 209

A.2.1 Output options 210
A.2.2 Tapping schedule 211
A.2.3 Chemical additions 212
A.2.4 Control variables 212
A.2.5 Current and voltage 213
A.2.6 Cell type 213
A.2.7 Model variables 213
A.2.8 Bath composition 214
A.2.9 Cell geometry 215
B Output from the Simulation Program 216
C Graphical Presentation of Results 218
References 220
Nomenclature 229
VII
List of Figures
Figure Following
number page
1.1 Cost of Production against Time 1
1.2 Schematic Diagram of the Hall Heroult Cell 4
1.3 Schematic Diagram of the Soderberg Cell 4
1.4 Anode Effect Termination Sequences 9
3.1 Schematic of the Current Efficiency Model 38
3.2 Cell Current Efficiency vs Alumina Concentration 44
3.3 Aluminium Concentration across CO 2 Boundary
Layer 49
3.4 Heat Flow and Temperature Gradient across Frozen
Sidewalls 51
3.5 Variation of Freeze Thickness with Parameter f 53
3.6 Local Nusselt Numbers as a Function of Local
Rayleigh Number 57
3.7 Fanning Factors for Determination of Effective
Anode Area 59
3.8 Schematic Diagram of the Model for Resistance
across the Bath 60
3.9 Models for Electrolyte Conductivity 62
3.10 Results of Fellners Simple Model for Binary Systems 63
3.11 Results of Fellners Simple Model of Tertiary System 63
4.1 Conceptual Model of the Aluminium Reduction Cell 75
4.2 Electrolyte Mass Balance Processes 81
4.3 Random Variation on Mass of Alumina Fed to a Cell 97
4.4 Schematic Representation of Reactions at an
Electrode 105
4.5 Schematic diagram of Tafel Derivation i+ »i- 106
VIII
Following
page
4.6 Overvoltages from Zuca 111
4.7 Overpotential vs Current Density 113
4.8 Overpotential vs Alumina Concentration 114
4.9 Overpotential by Mixed Model vs Alumina
Concentration 114
4.10 Current Density-Potential Curve for a Mixture of Two
Compounds Oxidising at Different Potentials 118
4.11 Schematic Diagram of Heat Flow in an Aluminium
Cell 121
4.12 Heat balance across the cathode 125
4.13 Heat balance across an anode 129
4.14 Cell Sidewall Profiles in Old and New Cells 138
4.15 Current Drawn by a Cold Anode 143
5.1 CElTROl distributed control system 156
5.2 Voltage Control Block Diagram 160
5.3 Anode Effect Termination Sequences 165
5.4 Cell Current Efficiency vs. Alumina Concentration 166
5.5 Target Voltage for Adaptive Feed Strategy for Point

Feed Cells 169
5.6 Typical Voltage Trace for Underfeed/Overfeed

Control Strategy Profile for 151 kA Centre Break Cell 170
6.1 Simulated Voltage Profile for 151 kA Centre Break

Cell 177
6.2 Simulated Voltage Profile for 151 kA Centre Break

Cell 177
6.3 Simulated Current Efficiency, Concentration and

ACD Profiles for 151 kA Centre Break Cell 179
6.4 Simulated Temperature Profile for 151 kA Centre

Break Cell 180
IX
Following
page
6.5 Simulated Ratio vs Temperature for 151 kA Centre
Break Cell 180
6.6 Simulated Bath and Pad Depths for 151 kA Centre
Break Cell 181
6.7 Alumina Concentration over First 24 Hours of the
'Standard' Run 182
6.8 Measured Variation of Alumina Content in Reduction
Cell Bath 182
6.9 Measured Variation of Alumina Content in Reduction
Cell Bath 182
6.10 Reduction Cell Operation with Two-Hour Break/Feed
Cycle 182
6.11 Simulation of Alumina Concentration, Freeze
Thickness and Cell Voltage 182
6.12 Simulated Effect of Anode Change upon Alumina
Concentration 183
6.13 Comparison between Simulated and Actual Voltage
Trace During Anode Effect in Cell 2A 18 183
6.14 Voltage Trace of Anode Effect Predicted by Model 184
6.15 Resistance Trace during Tapping of Pot 1A38 184
6.16 Change in Temperature and Bath Depth after a

Simulated Decrease in the Target Voltage of 0.1 V 184
6.17 Change in Temperature and Bath Depth after a

Simulated Increase in the Target Voltage of 0.2V 184
6.18 Simulated Effect of Anode Movement During Anode

Effect Termination 185
6.19 Simulated Dependence of Ratio upon Temperature 185
6.20 Measured Ratio in Anglesey Aluminium's Cell 1B61 185
x
Following
page
6.23 Typical Temperature-Time Curves for Two Different
Feed Modes 186
6.24 Simulated Temperature Profile over a 67 Minute
Feed Cycle 186
6.25 Average Temperature in an Industrial Cell After a
0.1V Increase in Operating Voltage 186
6.26 Temperature in an Industrial Cell During Anode
Effect 189
6.27 Simulated Temperature Profile During Anode Effect 189
6.28 Simulated Results of 24 Hour Tapping Schedule 190
7.1 Real Time Simulation in Parallel with Working Cell 193
7.2 Use of Model in Real Time Control 193
7.3 CElTROl Analog/Digital Inputs and Outputs 196
7.4 RS232 Link Between Simulation and CElTROl 196
7.5 Simulation Connected to Working Cell 196
7.6 Flow Diagram for DOS Task Scheduler 197
7.7 Inferential Control 198
7.8 Relationship Between Limiting Current Density and

Alumina Concentration 199
7.9 Electrolyte Probe for Reynolds Alumina

Concentration Meter 200
7.10 In Situ Alumina Concentration Meter 201
XI
List of Tables
Table Following
number page
1.1 Major Cost Centres in Aluminium Production 1
3.1 Viewpoints on the Rate Determining Step for
Alumina Dissolution 37
3.2 Current Efficiency vs Alumina Concentration:
Literature Data 45
3.3 Commonly Used Dimensionless Groups for
Determination of Heat Transfer Coefficients 54
4.1 Experimental Tafel Coefficients for Anodes of
Varying Porosity 111
4.2 External Resistances 121
4.3 Heat Transfer Coefficients in Aluminium Reduction

Cells 139
6.1 Comparison of Simulated Performance and Actual

Performance for a Half Break Cell in 1986 177
6.2 Aluminium Tapping Schedule at Anglesey Aluminium 178
6.3 Average Parameters From Five 170 Hour

Simulations of a Half Break Cell 191
XII
1.0 History of the Hall-Heroult Process
The electrolytic process for the industrial production of aluminium was
patented in 1886 by P.L.T Heroult in France and, independently, by C.M.
Hall in the U.S.A. The process involves the electrolysis of a molten
solution of alumina in cryolite and utilises the reaction
The carbon anode is oxidised during the reaction and aluminium is

produced at the cathode. The function of the cryolite is as a solvent for the
alumina and is required because of the high melting point of pure alumina.
The electrolytic production of aluminium increased from about 40 tonnes

in 1890 to 450 tonnes in 1895. The first commercial cells were operated
with a current of 4000 A and an energy consumption of 42 kWh per kg
aluminium. Even under these conditions, the price of aluminium was only
half that of the cheapest aluminium produced by the Ste.-Claire Deville
process which was based upon the chemical reduction of aluminium by
sodium. Improvements in cell construction were quickly made and by 1893
the energy consumption had already dropped to 25 kWh per kg
aluminium. The Hall-Heroult process had been shown to be commercially
viable and remains little changed to the present day.
1.1 Main Costs Of Production
Figure 1.1 shows how the cost of production and the price commanded for
aluminium has changed over the 55 years prior to 1985,[1] It is clear that
overall there has been a trend of decreasing real profitability since 1930.
This is not due to a similar trend in capital costs but rather to the increase
in world production especially due to the building of large smelting plants
in developing nations to utilise cheap electricity and labour.
For a smelter to remain in business it is necessary to reduce the

expenditure of all cost centres. The typical percentage cost of each stage
is shown in order of priority in table 1.1. The actual values may vary
significantly from smelter to smelter especially with regard to comparisons
between developed and developing countries.
1
1890 1910 1930 1950 1970
Figure 1. 1 Cost of Production against Time

0/0 of total cost / tonne of
Aluminium
Alumina 35
Electrical Energy 25
Anode Materials and Baking 12
Labour 10
Electrolyte Materials 3
Cathode Replacement 2
Table 1. 1 Major Cost Centres in Aluminium Production

The cost for replacement of the carbon cathode shown in table 1.1 is
based upon a cathode life of at least 3 years. If operation of the cell
results in the cathode life being much less than this, then replacement
costs can become highly significant.
It can be seen that the cost of the alumina fed to the cell constitutes the
largest percentage of the cost of producing aluminium. In the case of
Anglesey Aluminium, the alumina is purchased upon the open market
and, although alumina quality affects the operation of the cell, control over
alumina costs does not rest directly with the smelter.
The most significant area for improvement is therefore in the utilisation of

electrical energy. The other costs, anode, cathode and electrolyte
materials, are all inter related with energy costs. Minimum energy
consumption cannot be achieved without good quality anodes which
depends upon efficient baking. A well designed cathode and a properly
maintained electrolyte composition are also essential. The key area within
this section is the day to day operation and control strategy of the cell.
Increased computerisation has not only reduced the manual involvement
in cell operation, thus saving on labour costs, but has also allowed control
to be standardised on all cells within a smelter. The data logging facilities
have increased the amount of data collected for individual cells aiding in
the understanding of the process.
The capital costs involved preclude such methods of improvement of

energy utilisation as the rebuilding of the cell superstructure with modern
low resistance materials. With a computer control system already installed
in a smelter, as at Anglesey Aluminium, it is cheaper to make
improvements by changes in the control strategy for a cell. The aim of this
work is to produce a fully dynamic model of an aluminium production cell
to study the effects of different operating conditions and control strategies
upon the operation of the cell.
1. 1. 1 Fundamental relationships
There are three fundamental relationships that dictate the way in which an
aluminium cell should be operated. These are
2
increasing current and increasing current efficiency gives increased
productivity
I i + CE i ~ productivity i
increasing current efficiency and decreasing voltage gives decreased

energy usage
CE i + V .t ~ energy .t
increasing current density gives increased energy usage
cd i ~ energy i
In order to increase current for improved productivity without increasing

the energy consumption of the cell it is therefore necessary to maintain
the same current density by increasing the anode area.
The economic climate will dictate which of these relationships are the
most important. At times when aluminium can command a high price it is
desirable to increase productivity at the expense of energy efficiency,
while at times when the price of aluminium is low it is most important to
reduce production costs by increasing energy efficiency. Any control
strategy must be able to accommodate both these situations without the
need for major changes to the hardware.
1.2 Types of Cell

A commercial aluminium cell operates by passing a current through an
electrolyte consisting primarily of alumina dissolved in molten cryolite. The
ratio of the mass of sodium fluoride to aluminium fluoride in the electrolyte
is modified from the natural 1.5 of cryolite by the addition of excess
aluminium fluoride. Other additives, such as calcium and lithium fluorides
may be used to reduce the melting point of the electrolyte and to improve
the operability of the cell. The electrolyte, or "bath", is therefore a complex
mixture of molten salts.
3
The aluminium cell or IIpotll consists of a steel box lined with a refractory
brick. A carbon cathode block lies on top of this and forms the floor of the
cell. The cell is partially filled with the molten electrolyte and carbon
anodes are immersed in the electrolyte. A layer of molten aluminium, the
ll
IIpad forms at the surface of the cathode block protecting the carbon from
the corrosive action of the electrolyte and taking over the role of the
cathode.
The sidewalls of the cell are designed to allow a crust or IIledge ll of frozen
cryolite to form around the cell. This contains the molten electrolyte and is
very important in protecting the cell lining from corrosion and in
maintaining the thermal stability of the cell.
The only really fundamental technological change to the Hall-Heroult

process has been the development of the Soderberg anode. Unlike the
original cells which used an anode manufactured from coke and pitch
baked in a separate process, the Soderberg anode uses the heat
generated within the operating cell to bake these materials in situ. Coke
and pitch are fed in the top of the anode and by the time they reach the
reacting surface they have baked into solid carbon. The two types of cell
are shown schematically in figures 1.2 and 1.3.
The Soderberg anode was first patented in 1923 by the Norwegian

company Elektrokemisk and utilised a horizontal stub that entered the
anode from the sides. It was later shown that the fumes from the top of the
anode might be carcinogenic and the difficulty of containing them put the
Soderberg cell out of fashion. All new smelters since the middle of the
1970's have used the older 'pre-bake' anode technology. Research has
however continued on the Soderberg cell and methods of operation have
been developed that make this type of cell safe to use although it has not
gained widespread popularity.
As its name implies, the pre-bake anode is manufactured in a separate

process to that of electrolysis of aluminium. Whilst this may not be so
energy efficient, it does have advantages in that the anode quality may be
more easily controlled and poor anodes rejected before they are used in
the cell. An aluminium reduction cell contains banks of anodes made up of
individual units, in the case of an Anglesey Aluminium cell, two banks of
4
Fume Collection Anode Bus Bar
Alumina ~II-- _ - - - Anode Rods
Removeable Covers
/
Carbon Anode
Graphite Rammed
Sidewall Carbon
Current
Collector Bar
Thermal Insulation
Figure 1.2 Schematic Diagram of the Hall Heroult Cell

Anode Conductors
Steel Anode Studs
~// // // // // // // // // // // / /li:tJ I ~:~~~d
Thermal Insulation
Figure 1.3 Schematic Diagram of the Soderberg Cell

nine anodes making a total of eighteen in all. As the anode carbon is
consumed by the reaction, individual anodes must be changed at periodic
intervals.
Pre-bake cells are further classified by the method and frequency of

alumina feeding. The distinctions are made between Centre and Side
Break cells and between Point Feeders and Half Break cells.
The addition of alumina to a cell, necessitates the breaking of the surface

crust. In early cells, this was done manually and so the crust was broken
at the outer edges of the anodes, along the sides of the cell. This type of
cell is known as a side break cell. These cells are still used In some
smelters and automatic crust breakers have been developed.
The development of mechanical crust breakers for individual cells

operated by air pumps allowed for the positioning of crust breakers in the
centre of the cell. By feeding in the centre trench between the anodes, the
alumina falls directly into the bulk of the electrolyte and so can be
dissolved more rapidly giving some increase in the controllability of the
alumina concentration within the cell.
The feeding of cells manually and by simple timer mechanisms required

relatively large amounts of alumina fed infrequently. In centre break cells,
this led to the development of half breakers. These break an area of crust
in the centre trench approximately half the length of the cell. To maintain
an even distribution of alumina throughout the electrolyte, the cell is fed
alternately at either end.
The addition of alumina in large batches introduces large disturbances to

the cell and it is desirable to be able to maintain alumina concentrations at
an approximately constant level by continuous feeding. This can be
achieved by feeding only small amounts of alumina at frequent intervals. It
was not until the advent of digital computer control that this could be
successfully implemented. The resulting cell is termed a point feeder since
it only breaks a small area of crust in the centre trench, sufficient for the
addition of small amounts of alumina in the electrolyte.
5
1.3 Operation of a Prebake Cell
The day to day running of an aluminium smelting cell consists of a number

of operations that must be performed at different intervals. The actions
may be carried out either automatically or manually and the decision as to
which action to take may be based upon measured data, as in the case of
voltage control, or upon the experience of the cell room operator, as in the
case of voltage set point changes.
1.3. 1 Data Sampling
In order for decisions to be made on what actions to take in the operation

of an aluminium cell, information on the cell behaviour must be collected.
This consists of data obtained both by chemical analysis and by
observation by the operators of the general appearance of the cell. The
data can then be divided into two categories, one which can be measured
and one which is observed.
Measured data
Measurable data includes such information as electrolyte composition,
cell voltage and electrolyte temperature. These can be measured with
different degrees of accuracy and frequency. For example, cell voltage
can be measured almost continuously and with good accuracy. At
Anglesey Aluminium, the control computer, GELTROl, samples
individual cell voltages at a 100 ms interval to 4 significant figures
accuracy. Bath composition, on the other hand, is measured by mass
spectroscopy methods which are time consuming and cannot be
performed continuously. Although the accuracy of this method is good
for individual samples, it is not necessarily representative of the whole
cell.
Observed data
Appearance of the cell. eg flame colour, crust cover, movement of the
electrolyte. This type of data is subjective and depends upon the
experience of the operator for its usefulness.
6
1.3.2 Alumina feed
Alumina is fed to a cell by breaking the surface crust and allowing the
crust along with a mass of alumina to fall into the electrolyte. The alumina
on the surface is then replenished and a new crust allowed to form.
In early cells, the crust was manually broken using metal bars and alumina
was shovelled onto the surface. The process was first mechanised by the
development of large vehicular crust breakers that were driven along the
sides of the cells breaking the crust by means of a cogged wheel. Alumina
was replenished from a following truck.
Mechanisation of individual cells used mechanical crust breakers activated

by simple timers or, more recently, digital computers. At Anglesey
Aluminium, the breakers are suspended over the centre trench and are
operated by air motors. The motors are operated with compressed air that
is controlled by a solenoid activated either by a control computer or
manually. The motors control both the breaking of crust by movement of
the breaker beam and the dumping of alumina onto the surface by
operation of a gate in the alumina hopper suspended above the cell.
There are two sets of motors allowing individual feeding at both the tap
end and duct end of the cell.
1.3.3 Voltage control

The control of voltage drop across the aluminium reduction cells is by
movement of the anodes, thus altering the inter electrode gap. Continuous
voltage control has only been possible with the advent of computers that
can continuously monitor the cell resistance. The inter electrode gap is
adjusted by movement of the anode bus bar. Thus the anodes are moved
as a block and there is no facility for the automatic movement of individual
anodes. As with the feed system, the bus bars are moved by air motors
using compressed air. The solenoid switches may be activated either by
computer or manually with the anode movement being measured in time
rather than distance.
7
1.3.4 Anode effect termination
Anode effect occurs when the alumina concentration in the electrolyte falls
below a critical level. During anode effect, the composition of the bath
results in dewetting of the anodes and the formation of large bubbles on
the anode surface. The bubbles are predominately carbon tetrafluoride
and present a physical barrier to the passage of current into the
electrolyte. Anode effect cannot, therefore, be terminated by the simple
addition of alumina to the melt thus increasing the alumina concentration
above the critical level, but requires the breaking of the gas bubble layer
for alumina addition to be effective. This may be achieved by a number of
methods.
The first stage in anode effect termination is to agitate the electrolyte by

movement of the anodes. This works by both pumping the gas from under
the anodes and by bringing the anodes in contact with streamers of
aluminium from the metal pad. This causes momentary short circuiting
and allows the gas to disperse without being replenished by electrolysis.
Other methods of dispersal of the gas by anode movement have involved
the tilting of the anodes[21.
Should anode movement not be successful in terminating anode effect

then a green wood pole is pushed into the electrolyte. As the pole burns
violently the combustion gases cause a great deal of turbulence in the
electrolyte. This both physically breaks up the bubble and causes shorting
due to the throwing up of aluminium streamers from the pad allowing the
bubble to dissipate.
In Anglesey Aluminium's cells, operated under the GELTROl control

system, anode effects are terminated automatically. The computer
performs a sequence of the following actions:
raise anode
lower anode
break duct end
break tap end
feed duct end
feed tap end
8
A typical sequence for a half break cell is shown in figure 1.4. Actions may
be repeated a number of times by means of a loop construct or bypassed
on the condition of the anode effect being terminated by means of a jump.
These sequences are very effective with a 95% kill efficiency on point feed
cells. If the sequence fails in terminating anode effect, however, the
traditional method of using a green pole is still used.
1.3.5 Metal tapping
Molten aluminium metal is removed from the cell by a process known as

tapping. The process has changed little over the years and is still
performed manually.
In the early aluminium cells, aluminium was removed by simply breaking a

hole in the cell wall, allowing the molten metal to pour out and plugging
the hole with clay. This was hot and dangerous work and later methods
made use of compressed air to siphon the aluminium into a receptacle
known as a crucible.
Aluminium is still tapped by this method. A hole is broken in the surface

crust of the cell at the tap end and the spout of the crucible lowered into
the molten metal. The crucible is attached to the compressed air line on
the cell superstructure and aluminium is siphoned off until a desired
amount has been removed. The crucible is then moved on to the next cell
that requires tapping and the procedure is repeated. At Anglesey, one
crucible is used to tap four cells before it is taken across to metal servicing
to be formed into ingots. This amounts to a total mass of metal of about
4000 kg.
The method of measuring the aluminium collected remains crude,

although electronic scales have recently been introduced to the pot room.
The depth of aluminium in the cell is measured by means of a dipstick and
the difference between this measurement and the desired depth
determines the amount of metal to be tapped. Until recently, the metal
tapped into the crucible was also measured using a dipstick but the
introduction of electronic scales has seen a change over to tapping by
weight. This has seen some improvement in the accuracy of tapping as
the scales can be reset between each cell. Using the dipstick method, the
9
0 1 2 3 4 5 6 7 8 9
0 BRAK LOWR 7.0 W006 JUMP 15 RAIS 5.0 DUMP BRAK
1 LOWR 5.0 W006 LOPO 1 6 -RAIS 3.0 WOOS RAIS 3.0
2 W006 BRAK WOOS DUMP DUMP DUMP DUMP
3
4 WOOS BRAK W010
5 DUMP DUMP DUMP DUMP
Failure Location at: 15 Feed loop start at: 38

Plus seconds up at: 2 Failure feed seq at: 0
Acc. Feed Point at: 50 Max feed allowed at: 34
AVe on at: 38 Feed loop end at: 42
Sequence for Half Break Cell
0 1 2 3 4 5 6 7 98
0 DUMP BRAK LOWR 2.0 DUMP BRAK JUMP 13 LOWR 2.0
1 DUMP BRAK W001 LOPO 18 JUMP 24 RAIS 3.0 DUMP
2 BRAK LOWR 3.0 W001 LOP1 217 BRAK DUMP -RAIS 2.0
3 BRAK DUMP WOO2 RAIS 2.0 W002 RAIS 0.0 -W04S BRAK
4 DUMP DUMP
5

Sequence for Point Feed Cell
Nomenclature
BRAK activate crust breaker, both ends

DUMP activate alumina dump, both ends
LOWR n.n lower anode bus for n.n seconds
RAIS n.n raise anode bus for n.n seconds
JUMP nn jump to location nn if anode effect terminated
Wnnn pause nnn periods of 6 seconds = nnn x 6 seconds
LOPn xxyy loop to position yy allowed xx times if AE not terminated
e9 LOPO 1 6 loops to 6. Loop is allowed only once
decision point
Figure 1.4 Anode Effect Termination Sequences

aluminium tapped was measured cumulatively with the result that if one
cell was over tapped then subsequent cells would be under tapped.
1.3.6 Addition of raw materials other than alumina
As well as replenishing alumina being consumed by electrolysis it is

necessary to maintain the composition of the electrolyte by the addition of
other raw materials. The principal of these is aluminium fluoride which is
added to maintain the ratio of sodium fluoride to aluminium fluoride in the
bath. This ratio fluctuates for a number of reasons:
1. Sodium oxide in the alumina feed reacts with aluminium fluoride in the
melt to form sodium fluoride.
2. Aluminium fluoride is consumed by electrolysis during anode effect.
3. Both sodium fluoride and aluminium fluoride are volatilised off the bath
surface. This becomes more severe at higher temperatures.
4. The melting and freezing of the ledge causes changes in the ratio. This
is because the electrolyte freezes with a ratio closer to that of pure
cryolite at about 1.40. In an electrolyte with a ratio less than that of
cryolite this effectively means that sodium fluoride freezes preferentially
to aluminium fluoride with the result that freezing causes the ratio to
decrease and vice versa.
This tends to result in a ratio greater than that desired and it is necessary
to add aluminium fluoride to the bath. This is simply done by adding
bagged aluminium fluoride to the cell at a fixed interval. A typical example
would be 25 kg aluminium fluoride per day. The bag of aluminium fluoride
is simply thrown into the centre trench of the cell and broken into the
electrolyte by the action of the crust breakers.
The overall mass of electrolyte in the cell may also need replenishing. This
can be due to volatilisation of the electrolyte in a hot or poorly covered cell
or due to the splashing of bath out of the cell by anode movements. The
need to add electrolyte is not common in mature cells but is necessary in
cells shortly after start up. This is done by siphoning electrolyte from a
10
mature cell using the same method and crucible as for metal tapping and
blowing it into the receiving cell using compressed air.
1.3.7 Anode change
During electrolysis, the carbon anodes are consumed by the formation of

carbon dioxide. In a pre bake cell, it is therefore necessary to replace the
anodes periodically. The time between anode changes depends upon the
size of the anodes and the operational current of the cell. At Anglesey,
with the cells operating at 151 kA and with anode dimensions of 80x134x
80 cm, the 18 anodes in a cell are all replaced over a 21 day period.
Problems in cell operation or poor quality anodes may cause spiking, in
which an anode short circuits to the aluminium pad due to uneven
consumption of the anode, or burn offs, where poor covering of the top of
the anode allows surface oxidation around the metal stub allowing the
anode to break free and fall into the electrolyte, which necessitates un-
scheduled anode changes.
A spent anode is pulled from the cell by a small Icrane ' . If necessary, the
hole is then enlarged to take the new anode by drilling out the frozen
sidewall. The new cold anode is then placed in the electrolyte and bolted
to the bus bar. The positioning of the anode on the bus bar is adjusted to
give a uniform inter electrode gap within the cell. As the metal pad has a
humped profile due to the electromagnetic forces in the cell, the depth of
the anode relative to the bus bar is determined by its position within the
cell.
A new anode is much colder than the electrolyte into which it is being
placed, about 200 0
e compared to 970 o
e. A layer of frozen electrolyte
immediately forms around the anode effectively insulating it both
electrically and thermally. This frozen layer melts as the anode warms
over the next 30 hours but in order to increase the speed of melting it is
common practise to set a new anode low so that the frozen layer can
contact the metal pad. This prevents the freeze from growing and heat is
then conducted more speedily through the freeze and into the anode. On
the day following the anode change, the new anode is raised to the
correct position.
11
1.3.8 Sampling
Many of the parameters important to the operation of an aluminium cell

cannot be determined continuously or with great accuracy. Such
parameters are the bath and metal compositions, liquid levels, cathode
and ledge profiles and cell temperatures.
The bath and metal compositions are determined on a daily or weekly

basis by mass spectroscopy. Samples are removed from the cell in small
crucibles and processed in the laboratory. Impurity levels in the
aluminium, as well as being important for quality control can indicate
whether the aluminium is corroding the cathode or sidewalls leading to a
cell failure. The bath composition results can be strongly affected by the
sample point in the cell if, for example, the sample is taken close to a
sludge pocket. Alumina concentrations are not available from mass
spectroscopy but may be determined by chemical analysis. It is not
possible, however, to distinguish between dissolved and undissolved
alumina.
Liquid levels are determined on a daily basis by inserting an iron rod into
the bath and metal. When the rod is removed from the cell, the aluminium
and electrolyte freeze, the metal forming a thinner layer. The depth of
each section may then be read off against a rule on the side of the cell
and the values used to decide the amount of metal that should be tapped.
The corrosive environment of the electrolyte makes the use of permanent

thermocouples for the continuous measurement of temperature
unsuitable. Temperatures are measured on a daily basis using a hand
held thermocouple enclosed in an iron sheath. The sheath is slowly
dissolved by the bath and must be periodically replaced. The temperature
of a cell varies throughout the feed cycle and depends upon the position
of the thermocouple in the electrolyte. Temperature measurements taken
by this method are therefore more useful in indicating that the cell is
operating in the correct temperature range than giving an absolute
temperature measurement. Continuous temperature measurement has
been attempted by embedding thermocouples in the cathode structure
when the cell is built. This can give a better indication of temperature
fluctuations but has a number of drawbacks. An additional hardware
interface is required for each cell to monitor the thermocouples and this is
12
expensive. If a thermocouple fails, it cannot be replaced, and some means
of discarding the results from failed thermocouples is therefore necessary.
The determination of cathode and ledge profiles is labour intensive but

can be useful in evaluating the performance of a cell. The covers are
removed from the cell and holes punched through the crust between the
anodes and in the centre trench. Iron rods are then inserted in these holes
and the height of a reference marker is determined with a theodolite.
In the absense of high quality quantitative data regarding the operation of

a cell, an exception oriented process control system can be developed.
Such a system is described by S Manaktala [3] and is designed to assign
numerical severity values to visually monitored process exceptions in the
reduction cell. These exceptions are compiled as a potline survey and
action taken through specifically designed follow up procedures.
The severity indices for individual cells are stored and can be used In
conjunction with quantitative results to isolate and control chronic cell

operation problems.
The potline survey can be carried out by the line supervisor and cell
operators and includes an assessment of the flame coloration, pot voltage
and stability, carbon dusting in the bath and liquid levels. A severity index
of 0.5 to 3 is assigned to a cell having one or more abnormal attributes
and the required corrective action may then be taken.
References
1 Kent, J. H., "A Review of the Hall-Heroult Process", Aluminium and

Energy, A 100 Year Partnership Conference and Exhibition, Bangor,
Sept 18-19, 1986
2 Saksvikronnig, T., Valsvik, G., and Hove, S. J., "Anode Effect

Quenching by Anode Tilting", AIME Light Metals, 1982, p553-558
13
3 Manaktala, S., "Process Control Techniques for Reduction Cell
Operation", AIME Light Metals, 1971, pp165
14
2.0 Process Modelling
The usual method for evaluating the effect of changes in operating
parameters upon a system is by actual physical experimentation. This
type of experimentation can be very successful but its range is limited by
cost and safety. An alternative approach is to construct a mathematical
model of the process and experiment with the model instead of reality.
This has a number of advantages:
1) Understanding of the system
The development of the model demands an understanding of the process

to be gained. This insight enables extraneous "confusion factors" to be
stripped away from the problem allowing relationships between variables
to be seen more easily. Examples of confusion factors include changes in
raw materials or in ambient temperatures.
2) Economics
In industry, the cost of raw materials, labour etc can restrict extensive
experimentation. A suitably detailed process model however, is able to
quantitatively describe the behaviour of a system under a variety of
different operating conditions. Simulation studies can provide large
amounts of data at little cost and in a fraction of the time required for
experimentation. The cost of simulation of an experiment that results in
failure is no more than for one that is successful.
3) Optimisation and design
Simulation studies make it possible to study the effects of changes in

operational parameters easily and under reproducible conditions, which is
not always guaranteed in practise. Operation of the process can therefore
be optimised with respect to these parameters. The effect of changes in
design of the process may also be evaluated.
As fewer people need to be involved, especially at shop floor level, this

allows for the precise evaluation of designed operating procedure and
operational targets.
15
4) Safety
Mathematical models can be used to simulate process behaviour under

extreme conditions that cannot be practically or safely reproduced in the
laboratory.
5) Computer control
The advent of cheap powerful computers in the past few years has
permitted vastly more detailed and realistic analysis to be carried out for a
reasonable degree of effort and cost. More complex and fundamental
physical principles have been employed to formulate mathematical models
of processes. This has resulted in many processes being comprehensively
better monitored and controlled and has allowed control systems to be
designed without undue disturbance to the process.
6) Operator training
Process modelling makes possible the training of process operators using
simulation rather than real plant. Operators can gain first hand experience
in dealing with conditions that may occur only infrequently In process
operation as well as the implementation of safety procedures.
2.1 Model Development

The development of a process model will usually progress through a
number of stages:
1) Problem formulation and data acquisition

The first stage in model development is to decide upon the ultimate
requirements of the model. Following this, it necessary to obtain a detailed
knowledge of the complete system. This is usually achieved by using
literature data together with some experimental measurements. In
formulating a complete process model the complete process must be
studied, including such aspects as operator intervention. This last is best
investigated by spending a period on-site involved in the day to day
operation of the process.
16
2) Model description
Once comprehensive data on the system has been accumulated, the

process can be broken down into a number of relevant inter-reacting sub-
systems. From this a verbal description or conceptual model of the
complete process can be constructed.
3) Constructing the model
Each block in the conceptual model represents a system of algebraic and

differential equations. These equations are based on mass and energy
balances incorporating the appropriate kinetic expressions for the rates of
all relevant reactions and the rates of transport of mass and energy within
the system.
4) Model evaluation
Once the model has been developed it is necessary to evaluate it against

experimental data. If the models predictions do not compare favourably
under all operating conditions then it is unlikely to be correct. In this case,
either the model must be modified or a new model developed.
5) Application
If the model is able to accurately describe the system, it can then be used
to simulate process behaviour under a wide range of conditions. Any
assumptions made in development of the model must be taken into
consideration, however, as the model may not be valid if these do not
apply.
In practise, real problems are often ill defined and difficult to specify. Lack
of experimental data often makes it difficult to construct a model that
provides a balanced image of reality. Too simple a model may fail
because it does not adequately describe the process, while too complex a
model will contain a great number of unknown parameters and variables
and may not be mathematically tractable. The complexity of model
constructed therefore depends upon the accuracy required and the
application for which the model is intended.
17
2.2 Types of Model
The simplest form of process model is an empirical relationship between a

number of process variables. This type of model is usually developed by
regression analysis of experimental data and does not consider the
mechanisms of the process. These relationships can be extremely
accurate but are strictly only valid for conditions that lie within the range of
experimental data used in analysis. Empirical relationships cannot
adequately describe the dynamic effects of a complete process but are
invaluable in describing the individual blocks. For example, in an
aluminium reduction cell, the relationship between electrolyte density and
composition and temperature is well described by an empirical
relationship[1]. However, a similar equation relating cell current efficiency
to a number of process variables[2] is unsuitable for use in a dynamic
model as it is based upon average conditions in the cell over a period of
time and does not take account of the actual mechanism of current
efficiency loss.
2.2.1 Lumped parameter models
A lumped parameter model makes the assumption that process variables

are uniform throughout space. This allows the dynamics of a system to be
represented by ordinary, rather than partial, differential equations. A
classic example is the assumption that a stirred tank reactor is well mixed.
The temperature throughout the reactor can then be assumed uniform. All
resistance to heat transfer from the reactor to the jacket is 'lumped'
together in the stagnant film at the vessel wall. In many applications this
treatment is highly successful.
2.2.2 Distributed parameter models

In a distributed parameter model, consideration is taken of the variation of
process variables in space. This can allow for greater accuracy but the
number of calculations, and hence time, required for solution will also be
increased. The modelling of a tubular reactor provides a good example of
a distributed model. The composition and temperature of the reactants
flowing through the reactor vary both axially and radially. A complete
dynamic model requires the solution of partial differential equations in both
directions.
18
The complexity of distributed parameter models may be reduced by
reducing the model dimensionally. In the tubular reactor example, if flow
through the reactor is turbulent then the system may be assumed well
mixed in the radial direction. Only one set of differential equations,
representing conditions along the length of the reactor, need then be
solved. What is lost in neglecting the radial distribution may be gained in
the simplicity of the resulting model.
2.2.3 Static and dynamic models
The major distinction that must be made in describing process models is

that between static and dynamic models. A static model describes a
steady state system or system at a given instant with a defined average of
states. A dynamic model describes the changing states of the system
through time as the process is subject to internal and external
disturbances. The application for which the model is required determines
the type of model that should be used.
For example, static models have been developed for the cathode lining of
an aluminium cell. These models are complex and can accurately predict
the temperature profiles for a particular set of conditions. The model is
used to design cathode linings based upon average conditions expected
in operating the cell.
In practise, these conditions vary throughout time. It would be possible to

construct a dynamic model of the heat balance and use it to simulate the
changes in temperature profiles through time. However, the inter-
relationships of all factors involved are extremely complex and the
resulting model would provide little improvement over the present method
of cathode design.
Dynamic models on the other hand, often sacrifice high levels of detail
accorded by multi dimensional distributed models for the sake of both
mathematical tractability and computation time. An obvious example of
this approach is that used in classical control system analysis where linear
ordinary differential equations are used throughout. These equations are
often 'fitted ' to process responses to known disturbances. Although this
approach may provide sufficient detail for controller tuning, it provides little
data for other engineering design.
19
2.3 Choice of Model
A large model of a complete system, such as that of an aluminium

reduction cell, consists of many sub-models representing balance and rate
equations. The choice of model for a particular sub-system, e.g. dynamic,
static, lumped, distributed etc, is, however, more often the subjective
choice of the model builder and is partly based upon the experience and
skill in model building and partly upon the computing power available. The
recent advent of more powerful computers has allowed the accuracy and
complexity of dynamic models to be increased.
The following chapter describes some of the models available in literature

for the simulation of aluminium reduction cells.
References
1 Kvande, H. and Rorvik, H., "The Influence of Bath Density in Aluminium

Electrolysis", AIME Light Metals, 1985 pp671-678
2 Grjotheim, K., Haupin, W. E. and Welch, B. J., "Current Efficiency--

Relating Fundamental Studies to Practice", AIME Light Metals, 1985
pp679-694
20
3.0 Mathematical Models of Aluminium Reduction Cells
Process models may be divided into two main groups, static and dynamic
models. A static model represents a steady state system, or a system at a
given instant in time, for a defined average of states.
For example, many static models exist for the heat balance in a cell. In
fact the heat flows will vary with changes in temperature, instantaneous
cell voltage, electrolyte composition and depth, cell design and ambient
temperature. These models assume constant values for all such
parameters.
3.1 Static Models
3. 1. 1 Model of M. C. Richard [1]
M.e. Richard developed a model in 1975 to predetermine the effect of

operating parameters resulting from design changes to the anode and
cathode configurations. The model is essentially a steady state analysis of
the cell and was designed to define the Revere cell in the range 102 to
122 kA only. As the model was designed for use only in a narrow range on
either side of the existing operating conditions first order equations were
used to describe many of the physical effects. The model made no
attempt to model many non-steady state and human factors such as metal
tapping. Heat losses were calculated by treating the cell as a furnace and
fixing a desired operating temperature from which the losses could be
calculated by a steady state heat balance.
Whilst this simple approach is adequate for the investigation of the long
term effects of changes within some operational parameters it cannot be
used to describe the dynamic effect of a cell to frequent changes such as
alumina feed and anode effect.
3. 1.2 Model of B. Sulmont and G. Hudault [2]

Sulmont and Hudault developed a two dimensional thermoelectric model
for all categories of prebake pots. The model can be employed to
determine the equilibrium profile of frozen electrolyte, determine
temperatures and voltages at any point and calculate current densities in
21
both the electrolyte and metal pad and peripheral heat flows. The pot is
represented by a cross section of unit thickness and divided into a number
of elements. In the particular instance published, 1683 intersections were
necessary for the required accuracy. Temperatures and potentials are
determined at each intersect by iterative calculation from an arbitrary initial
solution.
This type of model is very accurate in the simulation of steady state

profiles but the number of equations involved make it unsuitable for use in
dynamic simulation. Its primary use is therefore in the design or
modification of pots.
3.2 Dynamic Models
3.2. 1 Model of S.R Blake [3]
S.R Blake developed a model for the variation of alumina concentration in

industrial Hall-Heroult cells. The model did not attempt a full solution of the
heat balance as this was outside the scope of the study. Variations in the
heat balance were accounted for by a random variation in the dissolution
rate of the alumina into solution. The model is based upon a simple, but
quantitative, picture of the mass transfer processes whereby alumina
passes from the solid to the dissolved state.
This model is quite successful at predicting the trends of alumina

concentration for different methods of feeding but is not complex enough
to investigate control strategies based upon the heat balance as well as
the mass balance. It also cannot predict the power usage of a cell and so
an optimum alumina concentration must be assumed.
3.2.2 Model of A. Ek and G. E. Fladmark [4]

A. Ek and G. E. Fladmark developed a model for calculating the dynamic
behaviour of a complete cell. The cell is divided into a number of coarse
elements for which the the energy and mass balances are formulated. The
effect of the chemical reactions, supply of materials to the bath and the
freezing and melting of the side freezes are included in the model. This
model cannot produce such accurate profiles across the cell as multi-
22
nodal models like that presented by Sulmont and Hudault but accurately
describes the working of the cell. The accuracy of this type of model
depends upon the accuracy of the individual assumptions and
expressions used within it.
3.2.3 Model of T. Hashimoto and H.lkeuchi [51
T. Hashimoto and H. Ikeuchi also developed a dynamic model of a

complete cell similar to that developed by Ek. In this case the cell was
divided into eight parts and partial differential equations with regard to
heat flux were solved by the finite difference equation. In addition to the
calculations in Ek's model the boundary layer between the melts and the
side freezes was treated as a diffusion layer giving a more accurate
estimation of bath composition in this section.
3.3 Models for the Dissolution of Alumina
Although the mechanism for the dissolution of alumina in cryolite is not

clear, much work has been done on the measurement of the dissolution
rate of alumina. The experiments have all been done on a lab scale and
the methods of adding alumina and measuring the dissolution rate have
varied considerably.
Thonstad et aH 61 introduced alumina into a well stirred cryolite melt at

1030 °C and measured the dissolution ie the time required for the
disappearance of alumina particles in the melt by a visual method.
Gerlach[71 put a tablet of alumina into a stirred melt at 1015 °C and
removed samples at regular intervals which were analysed for alumina
content. At the International Course on Process Metallurgy of Aluminium
in Trondheim[81, Thonstad cites four different types of experimental
technique
1. Vigorous stirring causing effective dispersion of alumina (Thonstad[61,

Winkhaus[91)
2. Moderate stirring allowing formation of agglomerates with the

electrolyte
23
3. Dissolution from a stagnant bottom layer of loose alumina or sintered
alumina (Gerlach[7])
4. Rotating discs of sintered alumina
Experiments have also been done on the effect of the alumina feeding
method. Maeda et a1l 10] did bench scale experiments using industrial
composition bath at 970 ce. Alumina was poured into the melt through a
funnel and the dissolution time measured visually. No stirring was done
and a large number of alumina samples were collected world wide.
The rate determining step for alumina dissolution in industrial cells has not
been clarified. According to the Thonstad[8] it may be
• mass transfer
• heat transfer
• chemical reaction
Mass transfer models fall into two categories
1. Zeroth order models - dissolution rate independent of alumina

concentration
2. First order models - dissolution rate proportional to alumina

concentration
3.3. 1 Zeroth order models

Both Gerlach[7] and Thonstad[6] postulate a zeroth order mechanism at low
concentrations of alumina of the type
(1 )
1
where r = rate of dissolution of alumina g cm- 2 min-
ko = reaction rate constant cm min- 1
E = activation energy kcal mole- 1
R = universal gas constant kcal mole- 1 K-1
To = reference temperature oK
T = reaction temperature oK
24
For concentrations greater than 3-6 wt% alumina they seem to agree with
a first order model. The accuracy of measurement by both visual and
chemical methods is not very great and it may be that at low
concentrations the driving force for dissolution is such that a first order
mechanism is involved. Other workers[9] have found first order correlations
at all concentrations.
3.3.2 First order models
A first order model for dissolution rate IS given by Asbjornsen[11] after

Kachanovskaya[12] as
(2)
where c* = saturation concentration of alumina in melt g cm- 3

c = concentration of alumina in melt g cm- 3
This is agreed with by a number of workers (Winkhaus[9], Jain[13]) and

Thonstad[6] and Gerlach[7] at higher concentrations. The experimental
work by Maeda et aH10] showed the dissolution rate of alumina to be
proportional to the concentration for the slow feeding method.
Much work has also been done on the dissolution of alumina from
sludge[14,15,16,17,18]. All workers agree that the rate of dissolution from
sludge is slower than the initial rate for dispersed alumina particles in
suspension. This would be expected as the specific area of particles in
suspension will be many times greater than that of sludge but may also be
due to sub cooling and the formation of agglomerates. The rate of
dissolution is also reduced by the presence of aluminium between the
sludge and bath forming a physical barrier. Thonstad[14] determined that
sludge dissolved by a first order reaction when exposed to bath with 4-5 %
alumina by weight at 1020 °e.
3.3.3 Heat transfer models

In heat transfer control models the rate of dissolution is determined by the
rate at which frozen cryolite melts from around clusters of particles of
alumina powder. A.N. Bagshaw et aH19] investigated the effect of operating
25
conditions on the dissolution of alumina by using a voltammetric technique
for the in-situ measurement of dissolved alumina. Alumina was fed to
molten cryolite in a crucible by a variety of different methods and the
alumina concentration measured at five second intervals. The
temperatures of the bath and the crucible were also monitored. In many
cases a two stage dissolution pattern was observed, indicative of fast
initial dissolution followed by the formation, melting and dissolution of
agglomerates. The time for remelting of the agglomerates seen as a delay
in the dissolution of alumina can be calculated from
(3)
where hb = heat transfer coefficient from bath to agglomerate

rAI2 0 3 = radius of AI 2 0 3 agglomerate
Tb = bath temperature
TI = bath liquidus temperature
tremelt = time for remelt of agglomerate
Mf = mass of bath frozen into agglomerate
~Hfus = latent heat of fusion of bath
The right hand side of the equation is given as
(4)
where (MCp) AI 2 0 3 = mass of alumina x specific heat capacity of

alumina
The time for remelt is highly dependent upon the mass of the agglomerate
and its position in the cell. Bagshaw has calculated times of 4-6 seconds
for individual particles exposed to turbulent bath with high heat transfer
coefficient and 10°C superheat. Under the same conditions a much larger
aggregate of 1.2 g alumina would take 230 seconds to melt. However, in a
stagnant zone with a consequently lower heat transfer coefficient this time
would be doubled. The circulation of bath would also be required to break
up the agglomerate to prevent further freezing.
26
These experiments show the validity of a mUlti-stage model. The freezing
effect may be simplified by assuming a lower specific area of alumina in
suspension than the powder to account for the presence of agglomerates.
This approach would reduce the accuracy of a multi-nodal model but will
be adequate for a well mixed model in which an average heat transfer
coefficient will be assumed throughout the bath zone. This negates the
necessity of determining the sizes and flow patterns of agglomerates.
3.3.4 Addition of alumina to electrolyte.
It has been assumed by most modellers that on introduction to the bath

the alumina goes into a number of different states. This is borne out by a
number of experimental studies where the type of alumina and the method
of feeding have been found to effect the dissolution rate and formation of
sludge. Many models( Ek[4], Hashimoto[5]) assume alumina to exist in the
electrolyte in two states, dissolved and undissolved. A proportion of
alumina is assumed to go directly into solution and the rest into a buffer
from which it dissolves according to a first order model. This assumption is
likely to be valid when alumina is added to a cell with an alumina
concentration well below the saturation concentration. In this case the
concentration gradient would be great and the rate of dissolution very fast.
However this will not simulate accurately abnormal conditions such as the
effect of over feeding a pot unless the proportion dissolving instantly is
made a function of the concentration gradient within the electrolyte.
Blake[3] and Entner et aH20] assume three states of existence. These are
as dissolved alumina, dispersed alumina or suspension and bottom
sludge. Entner assumes that alumina is first dispersed into suspension
from which it dissolves by a first order model dependent upon the mass of
alumina in suspension, the temperature of the electrolyte and the time
since anode effect. The latter factor is to account for the enhanced
dissolution of alumina due to the perturbation of the bath during anode
effect. Alumina in suspension also sediments out to form a bottom sludge
from which it dissolves into solution. The rate at which it dissolves from
sludge is unclear but appears to have the same form as dissolution from
suspension.
Blake also defines the same three states as Entner but a proportion of the
alumina introduced is added to each of these states. Mass transfer is
27
assumed to occur sequentially from sludge to suspension and from
suspension to solution. Alumina sludge is transferred to suspension
according to the relationship
(5)
where R1 = the rate of transfer of alumina from sludge to

suspension
k1 is a function of the average bath velocity
M1 = the mass of alumina as sludge
K2 represents the sludge geometry. This final term allows
for the simulation of different designs of cell and feed
strategies.
Alumina dissolves from suspension by a first order model as above.
Blake1s model was primarily designed for a side break cell in which
feeding takes place mainly from the sludge. In centre break cells where
the amount of sludge should not be so great and alumina crusts may form
at the metal/bath interface, the rate at which alumina can dissolve directly
from sludge may be greater than the dispersion of alumina in suspension
by the movement of electrolyte over the sludge. This would account for
the difference between the models of Blake and Entner.
3.4 Models for the Evolution of Fluorine from Reduction Cells
Fluoride is lost from the cell during normal operation as well as by the
production of fluorocarbons during anode effect. The fluoride is often
classified as gaseous and particulate. Gaseous fluorides are those that
continue to be gases at ambient temperature, namely HF, SiF4 and
fluorocarbons. Entrained and volatilised electrolyte becomes particulate
on cooling. The exhaust gases are captured by the cells hood and passed
through the baghouse where a proportion of the fluoride is trapped and
recycled to the cells with the feed. Haupin[21J developed a mathematical
model for the prediction of fluoride evolution from Hall-Heroult cells using
thermodynamics, kinetics and the physical properties of the electrolyte.
28
3.4. 1 Model of W.E. Haupin
Haupin derives relationships for three types of evolution of fluoride;

gaseous, particulate by volatilisation and particulate by entrainment of
bath.
3.4.2 Gaseous fluoride
The major gaseous fluorides are HF and fluorocarbons. The loss of

fluoride as fluorocarbons can be calculated from the rate of reaction of
anode effect. The evolution of fluorocarbons during anode effect is not
considered in Haupin's model.
Haupin assumes HF to be the major gaseous fluoride evolved from the

bath during normal operation. This is generated by hydrolysis of the
molten electrolyte, water being introduced to the bath by
1. the water content of the alumina

2. burning of hydrocarbons from the anodes
3. moisture in the anodes
4. moisture in other bath additions
Fortunately only a small percentage of this moisture is converted to HF,

measurements by Alcoa indicating that only some 8.7% of this moisture
actually reacted.
The water is assumed to react with AIF3 in the bath by the reaction
[1 ]
since the equilibrium constant for H2 0 reacting with AIF3 is 103 times
greater that for H2 0 reacting with Na3AIF6 and 108 times greater than that
for H2 0 reacting with NaF.
The partial pressure of HF gas for this reaction is given by
(6)
29
where PHF = partial pressure of HF gas
L\G = Gibbs free energy of reaction
R = universal gas constant
T = temperature of electrolyte
PH 20 = partial pressure of H2 0 in pot gas
aAIF3 = activity of AIF3 in electrolyte
aAI20 3 = activity of AI 2 0 3 in electrolyte
Expressing L\G as a linear function of temperature and inserting

appropriate literature data for activity coefficients into this equation, a
good match was obtained to Alcoa test data when it was assumed that the
reaction proceeded between 26.6 and 28.8% of the way to completion.
This resulted in the following equation for the evolution of gaseous fluoride
F = 2310000
G %CE
_ 955000 oR o[
P
( 8444JJ
exp 7.553- T
b 0
%H 20 in AI 20 3 + O/oH in anodesJ12 0
[ 37.44 21 .. 5
c ]-0.462
exp( 0.44199 - 3.1733R b + O. 78127R~) C:~~03 (7)
0
[ AI 20 3
where = gaseous fluoride evolved per metric ton aluminium

produced kg FI tonne AI
%CE = current efficiency as percentage
Rb = weight ratio NaF/AIF3
Pb = barometric pressure kPa
CAI203 = concentration of alumina in bath wtO/o
C~~;03 = saturation concentration of alumina wt°jo
This model shows a very good fit to the Alcoa measurements at

concentrations of alumina above 3 wt°jo. Below this value however the
prediction diverges rapidly from the measured relationship. One reason for
this could be that the measured relationship was derived by multiple
regression analysis with a linear relationship assumed between fluoride
evolution and each variable. The comparison is therefore being made
30
between the model and an extrapolation of a relationship derived by linear
regression.
3.4.3 Particulate fluoride by volatilisation
It has been demonstrated that the vapour over the electrolyte is NaAIF4
with a small but significant amount of dimerisation[22]
[2]
The mass spectrographic measurements of the equilibrium constant can

be correlated
Kd = -P2
d
= exp (21414
- +15.6 J (8)
Pm T
where Kd = equilibrium coefficient for dimerisation

Pd = partial pressure of Na2AI2Fa kPa
Pm = partial pressure of NaAIF4 kPa
T = temperature OK
From the data of a number of investigators Haupin has derived the

following formula for vapour pressure over electrolyte
(9)
where
2
A = 7101.6 + 3069.7· Ratio + 635.77 . Ratio + 51.22· (%UF)
764.5·(0/oAI 2 0 3 ) (0 C F )
-24 638·Ratio·(0/oUF) + )] + 13.2· Yo a 2
. [1 + 3.2029·(0/oAI 2 0 3
3
B = 7.0184 + 0.6844· Ratio + 0.08464 . Ratio + 0.01085· (%UF)
1.1385· (0/oAI 2 0 3 ) (0/ C F )
_ 0 005489· Ratio· (%UF) + )] + 0.0068· /0 a 2
. [1 + 3.2029· (0/oAI 2 0 3
0/0 = wt% of species in electrolyte
31
Assuming that the pot gas is saturated with NaAIF4 and Na2AI2 Fa at their
calculated vapour pressures Haupin arrives at the following equation for
the evolution of particulate fluoride
F - 204000 (Pm + Pd )
Vp - %CE· P kg F I tonne AI (10)
b
where %CE = current efficiency of cell 0/0
The calculated value of particulate fluoride by this method was always less
than that measured. The difference between the calculated and measured
values was assumed to be entrained electrolyte and was correlated
against bath composition and temperature.
3.4.4 Particulate fluoride by entrainment
Although some investigators have described mechanisms for the

entrainment of electrolyte[23,24,25] there is little published data on the
subject. Evidence of significant entrainment include the fact that the
NaF/AIF3 weight ratio of the particulate fluoride is intermediate between
that of bath vapour (NaAIF 4) with a ratio of 0.5 and that of the electrolyte
itself. The presence of calcium fluoride in the particulate also suggests
entrainment since calcium species in the bath have a negligible vapour
pressure.
The differences between the calculated and measured particulate was

assumed therefore to be due to entrained electrolyte. The data was
correlated empirically with bath ratio, alumina concentration and
temperature by multiple regression analysis and the following relationship
obtained
[-17030 + 29800· Rb -13000· R~ + 67· (0/oAI 2 0 3 )
-173·t -0.389·t 2 + 141.6·R b .t]

FE=-----------------~-o-C-E----~----------- (11 )
where FE = Entrained fluoride kg FI tonne AI

t = T - 1243 oK
32
Being derived by difference this correlation contains all the errors of
measurement and the predictive error of the other models. Additionally it
predicts negative entrainments for electrolytes with bath ratios above 1.4
at temperatures greater than 990 ae. This demonstrates the danger of
extrapolating an empirical correlation.
The total fluoride evolution is given by the sum of the gaseous and
particulate evolution
(12)
This is the total fluoride evolved from the cell into the exhaust gases.
These gases are passed to scrubbers where much of the fluoride is
recaptured and recycled. The total fluoride irrevocably lost from the cell is
therefore given by
(13)
where: Flos t = Fluoride irrevocably lost from the cell kg F/tonne AI

HE = Hooding efficiency of pot
GE = Scrubber efficiency for gaseous species
PE = Scrubber efficiency for particulate species
3.5 Models for the Loss of Current Efficiency
The current efficiency of an aluminium reduction cell is defined as the ratio

of the actual production of aluminium to the theoretical production of
aluminium by the current passing through the cell. In their review of
current efficiency modelling, Grjotheim et al[261, state that the deviation
from 100% current efficiency can be caused by all or one of the following
1) Other materials being co-deposited. This would be due to the

reduction of other metal ions in the melt such as sodium, calcium, iron
etc. This results in the presence of impurities in the aluminium.
2) Some of the product reduced at the cathode may remain In the

electrolyte phase and be subsequently reoxidised.
33
3) Electronic conductivity of the electrolyte and physical short circuiting
reduces the actual current used for electrolysis.
4) The formation of other heterogeneous compounds. Such an effect is

the formation of aluminium carbide at the cathode by reaction between
the aluminium pad and the cathode material.
5) Metal being removed from the product before the amount formed can
be assessed. Aluminium could be reoxidised in the electrolyte either
by dissolution from the metal pad or as a dispersion of droplets.
Obviously the effect of some of these mechanisms will be small and may
be ignored in the modelling of the process. The purity of the metal
produced is typically greater than 99.8% indicating that process (1) has
only a small effect on current efficiency.
Process (2) is likely to be important through impurities in electrolytes that

exhibit stable multiple valance states such as phosphorus and
vanadium[27]. These form reduced species in the electrolyte and are
cyclically reoxidised at the anode. This reduces current efficiency by
providing competing reactions at the cathode. The importance of this
effect will depend upon the concentration of these species in the
electrolyte. If it can be assumed that the amount remains constant, ie
these species are not present in the feed, then the effect can be modelled
as a constant reduction in the current efficiency of the cell.
The dissolution of a metal in a molten salt imparts an electronic

conductivity and lowers the current available for electrolysis. Physical
shorting of the anode and cathode due to poorly set anodes will have the
same effect. Although it is difficult to assess the loss due to electronic
conductivity it is clear that the effect is small. Physical shorting is an
exception to the normal running of the cell and is generally of a short time
duration. It could, however, represent a significant loss particularly if the
cell is to be operated at short anode to cathode distances.
The formation of aluminium carbide is the major reaction accounting for

process (4). The effect on current efficiency is self limiting due to the slow
diffusion of reactants through the carbide layer and is therefore limited to
34
the rate of dissolution of aluminium carbide in the metal and electrolyte
with which it comes in contact. Although it has a significant solubility in the
electrolyte, its solubility in metal is very low. Under normal conditions the
cell sidewalls are protected by a ledge of frozen electrolyte and carbide on
the cathode surface is protected from the bath by the metal pad. The
effect upon current efficiency is therefore limited by the rate of dissolution
into the metal and is very small. If the sidewalls become exposed, as after
an extended anode effect, then aluminium carbide will form and dissolve
causing erosion of the sidewall and loss of current efficiency until the
frozen ledge is once more established. These effects are of great
significance during the very early life of a cell or in a sick pot, but can be
considered negligible during normal operation.
The effects of all the above processes are fairly small and it is universally
accepted that the irreversible removal of metal by process (5) is the
primary loss mechanism. It occurs through the dissolution of one or both
of the electrode products and transportation to a reaction zone.
The most common method for the measurement of current efficiency is to

compare the metal produced by a cell over a long period of time, typically
a month or more, with the theoretical production based upon the average
current during this period. This method is used to assess the current
efficiency of industrial cells on a quarterly and yearly basis but is not
suitable for current efficiency analysis over a short time scale.
The alternative method of measurement is by analysis of the pot gases. In

1947 Waddington and Pearson[28] discovered the following relationship for
the ratio of CO 2 to CO in the anode gases and cell current efficiency
+ 50 (14)
This relationship allows instantaneous current efficiencies to be measured

although it has not so far been possible to monitor this ratio continuously
on a plant wide basis. The method can be used to determine the
instantaneous effects of operational parameters on current efficiency. The
correlation demonstrates that CO 2 is the primary oxidant and CO the
35
byproduct of the oxidation reaction. The metal reoxidation reaction IS
usually depicted
AI(soln) + 1.5C02 = 0.5AI 20 3 (soln) + 1.5CO(g) [3]
The reaction may be assumed to occur between CO 2 as a gas or in

solution. Studies of CO 2 solubility in the electrolyte[29] indicate that it is an
order of magnitude lower than then solubility of aluminium. For this reason
mechanistic explanations have assumed that the reaction was dominated
by gaseous rather than dissolved carbon dioxide. The interfacial area
between gas bubbles and electrolyte is much greater than that between
the metal pad and electrolyte and despite the much lower solubility of
carbon dioxide it is probable that dissolved carbon dioxide plays an
important role in the reoxidation mechanism. The interfacial area of carbon
dioxide as a function of alumina concentration determined by the
observation of gas bubbles in an experimental cell has been published by
Haupin[30]. The relationship is given by
0.02723 + 1.003· wt - 0.3208· wt

2
- 0.05874· wt 3 J
( +0.03268·wt 4 -0.001817·wt 5 xl
(15)
(1-0.4715.wt+0.06247.wt 2 ) 0
where Ac0 2 = interfacial area of CO 2 gas/electrolyte
wt = wt% of alumina in solution

10 = current associated with the oxide reaction
Unfortunately, no relationship has been derived for the dependence of

bubble surface area upon bath composition, anode properties and
temperature.
Laboratory studies of the reoxidation reaction have generally been aimed

at evaluating the reaction mechanism, especially the rate determining
steps. In Grjotheim's review[26] he states that there is a lack of agreement
as to the mechanism. However, under conditions similar to those found in
operating cells mass transfer appears to play an important role although
the rate determining step is not clear. There are several consecutive steps
for the dissolution and loss of aluminium. These have been documented
by Qiu [31] as being 3 steps
36
1. Dissolution-diffusion step. The dissolution of metal at the metal-
electrolyte interface and the diffusion of the dissolved species through
a stagnant cathodic layer. As the cathodic layer is usually very thin
these two steps may be considered together.
2. Transfer of the dissolved metal through the bulk of the electrolyte to a

reaction zone. Convective transport limited.
3. Reoxidation of the dissolved species by anode gases or dissolved

anode gases. Reaction rate limited.
Qiu also lists various viewpoints presented in the literature for the rate
determining step. This is duplicated in table 3.1.
These widely different conclusions have been explained by Grjotheim et

aH26] as possibly being due to the differences in the experimental
conditions and the way in which convection in the melts has been
achieved ie. by means of a temperature gradient, mechanical stirring or
gas bubbling. Where the melt is vigorously mixed the reaction is
dissolution rate controlled whilst convective transport is the slow step in
quiet melts. In industrial cells a combination of these steps may be
responsible, dependent upon local conditions within the melt.
3.5. 1 Empirical relationships
A number of investigators have correlated current efficiency with

operational characteristics for reduction cells[26,32,33,34,35]. These
correlations are generally obtained by multiple regression analysis of
current efficiency measurements for a number of cells, the operational
parameters being noted at the time of measurement. The current
efficiencies may be obtained by metal inventory measurements or anode
gas analysis techniques. An example correlation is that derived by Berge
et aH36]
CE =O.1388·T+O.59,X A'F3 +58.9sin(3h)-O.032·A+163.7 (16)
where CE = current efficiency %
T = bath temperature °C
XA1F3 = excess AIF3 in the electrolyte wt%
37
authors year experimental method rate determining
step
laboratory cells
Abramov 1953 determine the metal loss in convective

the bath with different depths transport
Revazyan 1960 determine the metal loss of dissolution

Cu-AI alloy in the bath
determine the metal loss in
Gjerstad
and 1964 the bath with different convective
depths, by passing CO 2 over transport
Welch
the bath surface.
determine the rate of
Gerlach 1974 oxidation of metal by passing
dissolution
CO? over the bath surface
Thonstad
1981 passing CO 2 over cryolite dissolution
and
melts in contact with metal reaction
Rolseth
Qiu an electroanalytical method
convective
and 1984 to assess the metal
transport
Fan concentration in bath
Industrial cells
determine the metal
convective
Arthur 1974 concentration in different
transport
zones of bath
an electroanalytical method
Thonstad
1976 to assess the metal reaction
and
concentration in different
Rolseth
zones in the bath
Qiu determine the metal
and 1984 concentration at different dissolution
Fan zones in the bath
Table 3. 1 Viewpoints on the Rate Determining Step for Alumina

Dissolution
h = depth of metal pad cm
A = cell age months
Current efficiencies were obtained from metal inventory measurements by

an isotope dilution method on a pot line of 34 cells. In the bulk of the cells
one value of current efficiency was determined each week. The operating
parameters must needs be the weekly averages for each cell.
Whilst multiple regression models can be used to predict average current

efficiencies under target or average operating conditions they cannot
provide the detailed information required for a dynamic model in which
operating conditions are constantly changing.
3.5.2 Model of Robl et al

Robl et aH371 presented a model for the calculation of steady state current
efficiency in terms of such geometrical and operating factors as anode-
cathode distance, bath temperature, current density and bath composition.
Hydrodynamic parameters are also included in the model.
The model is based upon the assumption that the reoxidation reaction
occurs between dissolved CO 2 and aluminium dissolved or suspended in
the bath. The two products are brought together by diffusion and mass
transport in the bath. The model is illustrated by figure 3.1. The anode to
cathode gap is divided into five regions
1. metal dissolution zone assumed to have constant metal concentration

less than or equal to the metal solubility in the bath
2. convective transport zone for metal assumed to have constant

gradient of metal concentration
3. reaction zone with overlap of metal and gas concentration gradients
4. convective transport zone for dissolved gas assumed to have constant

gradient of gas concentration
5. gas dissolution zone assumed to have constant gas concentration less

than or equal to the gas solubility in the bath
38
o
oOcfooo
00 000 0
0o~ o0
000
0
0
o 00000 0 0
Q..,: 0 0
~UOOOOOO
00 0 00
o gOo 0 0
0000000 0 0
o 0
On-O
0"V0000 0 0
00
0
'0 0000 0 OJ
o ~ °O'b
00
~O0~0?:''d0000Cb~00'cX)00CfOo~62-'
0 0
o 0 "0 0 '<3 0 0
0 :l, ';;0 'I!
000.'l.J'% 00 00
o~goo
0
~ifo?~oo °00~8000000'tJ00
_'I>oooo,!;/ 0 0 0
O~OO~ <Y. ~ 5fbQl,.e.,c'i"'rans

C port by~~ '()"CO- o.a 0
00
oon 0 00
'..r\';QQ':.o( '0_
0
,-0 00 00
0
0
000
0
Oooo 00 00
0
o o 0 0 00 0 00 0 000000000000 0 0
0~n~,ectoQ
0 0 v-o ..'cf: • 2 fO 0 0 00 dO80
00 000
0 0 0
0 0 00
o Reaction 0 c ...... cGradient 0 0 0
0
00
0
0 - !O 0 o 00 0 0 0
0
0ic'-J
0 0
0000
00000
00
0
00000
0 00 0 0
0
0 0
0
Metal Transport e 0 0 0 0 0 0 0
by Convection
Boundary Layer,
Metal transport
by Diffusion p / / / / / ~/ / / / / '/ / / / /
/ / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / ,
/ / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / ,
/ / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / ,
/ / / / / / / / / / / / / Aluminium Pad · / / / / / / / / / / / / / / ,
/ / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / ,
,
/ / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / /
/ / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / ,
Figure 3. 1 Schematic of the Current Efficiency Model

A hydrodynamic model of the cell was developed to allow the calculation
of differing current efficiencies in different zones of the cell since
distinctive flow regimes may be identified throughout the bath. The
turbulent flow pattern of the bath within the inter electrode gap is likened
to that in a channel whose walls are the anode and metal surfaces. It can
be assumed that a thin boundary layer exists at both the metal and gas
interfaces. In these boundary layers mass transport is by diffusion only,
the boundary layers being small compared to the depth of the channel.
Using an equivalent convective diffusivity as defined by
= °
m (17)
where Oem = equivalent convective diffusivity for metal species

Om = mean molecular diffusivity of metal species in bath
ut = turbulent viscosity of the electrolyte
SC t = turbulent Schmidt number
the rate of convective transport can be calculated by Fick's Law
dC
- -0 . - (18)
m dz
where rm = rate of convective transport of metal

Om = diffusivity of metal through the boundary layer
dC = concentration gradient across boundary.

dz
The turbulent Schmidt number may be related to the Schmidt and

Reynolds numbers but the calculation of turbulent viscosity is more
difficult. This is dependent upon the shear velocity, the friction factor, the
skewness of the velocity profiles and surface roughness. Robl derived the
following equation for equivalent diffusivity
[O.42.CI·~·Dm ·U .p2d~ (1-C s .~)]

2
(19)
+ [0.9. Om . U· p. de +.u 2 ]
39
where Jl = viscosity of the electrolyte
P = density of the electrolyte
U = average velocity of electrolyte reference to metal
d = interelectrode distance
de = critical dimension for Reynolds number
Z = distance normal to interface
Cf = friction factor
Cs = factor for skewness of velocity profile due to unequal
surface roughness
In the steady state, if the reaction between the two species is assumed to
be instantaneous then the rate of metal loss is equal to the rate at which
metal is transported to the reaction zone. The current efficiency is then
calculated as
(20)
where rt = theoretical maximum production of aluminium

I
---
ZAI· F
= current flow through cell
zAI = charge number for aluminium
F = Faradays constant
If the rate of convective transport is greater than the dissolution rate of

metal into the electrolyte then it wi" become dissolution rate limited
otherwise it wi" be equal to the convective transport rate. No mathematical
model is presented for the dissolution rate although it is stated that an
analytical expression for this should include the relative turbulences and
surface tensions of the bath and metal pad.
Although this model can predict current efficiencies in differing zones of

the cell many of the factors in equation (19), such as Cf and CS1 are
difficult to determine. Whilst metal pad velocity is relatively easy to
measure in industrial cells this is not the case for the electrolyte velocity.
The accuracy of the hydrodynamic model also depends greatly upon the
regional division of the metal and bath flow patterns.
40
An advantage of the model is that it is not restricted to the assumption of
any particular rate determining step. However, for known conditions in
certain Alcoa cells, the model without dissolution rate limits applied
predicted current efficiencies that were too low. Robl concluded that in
many instances it is likely that the dissolution rates for metal and/or gas
are the limiting factors.
3.5.3 Model of Evans et al
Evans[38] assumed that the rate controlling step for the reoxidation
reaction is transport through a surface tension induced boundary layer at
the aluminium electrolyte interface. Using a relationship derived by
Levich[39] they correlated the rate of transport through the interface to the
turbulent kinetic energy. The kinetic energy was obtained from their k-e
model used to calculate fluid flow. This lead to the relationship
O.5
rm - (D~~P J . C· .k~75 (21 )
where rm = rate of metal transport through boundary layer

Om = molecular diffusion coefficient of dissolved metal
p = electrolyte density
am = surface tension at aluminium surface
C* = saturation concentration of metal in electrolyte
kt = turbulent kinetic energy of the system
k is obtained by averaging the k-distribution obtained in their velocity

t
calculations. In calculating the electrolyte velocity gas bubble induced
forces are ignored, electromagnetic forces only being considered. The
steady state current efficiency is again calculated as
CE (22)
As with Robl's current efficiency model this includes a hydrodynamic

model. Although this may increase the accuracy and allow the calculation
of local current efficiencies, the number of calculations required make it
cumbersome for dynamic simulation and the need to make
41
approximations for many of the unknown parameters reduces these
benefits.
3.5.4 Model of B. Lillebuen et aJ
Like the previous models Liliebuen[401 assumed that the reaction between
aluminium and carbon dioxide was solely responsible for current efficiency
losses. In addition he makes the following assumptions:
1) the reaction takes place between dissolved reagants
2) simple film theory is valid and transportation of the heat of reaction and
the reaction products is disregarded
3) the chemical reaction is instantaneous and the rate of reaction is

considered to be controlled by the dissolution of reactants through
stagnant films
The dissolution rates for aluminium and carbon dioxide taking into account
the reaction can be written
rAI - AAI·kAI {1 + (23)
(24)
where r = rate of dissolution of species

A = interfacial area of species and electrolyte
k = dissolution rate constant
C = concentration of species
C* = saturation concentration of species
suffix AI refers to aluminium
CO 2 refers to carbon dioxide
An implication of assumption (3) is that if CA1>0 then Ceo2 =0 and vice

versa Ceo >0 requires CA1=0. In line with experimental data[41J Lillebuen
assumes t~at there is always dissolved metal present in the electrolyte.
42
As in the previous models the steady state current efficiency is given by
CE = 100{1- r;., J (25)
The stoichiometry of the reaction requires that
r.CO 2 -
-
3 r
2" AI (26)
combining equations (23), (24) and (26) gives
kAI . AAI . C~I - %. kC02 . AC0 2 . C~02

(27)
%. kAI . Aco 2 + kco2 . AAI
Calculation of the steady state concentration of aluminium in the

electrolyte allows the calculation of the rate of dissolution and hence the
current efficiency.
3.5.5 Calculation of mass transfer coefficients
Attempts to calculate the mass transfer coefficient for gas bubbles, kC02 ,
by correlations proved unsuccessful due to lack of experimental data for
CO 2 diffusivity and bubble diameters. Lillebuen therefore assumes that
kco = kAI based upon the work of Aarebrot[42] with the ratio of gas bubble
2
area to aluminium area being estimated as 0.1.
The mass transfer coefficient for aluminium is calculated from
Sh = 0.023· Reo.83·Sco.33 (28)
k ·2£
where Sh = Sherwood number = _AI__
DA'
p·v·2f
Re = Reynolds number - ~--
11
11
Sc = Schmidt number
£ = interelectrode gap
p = density of electrolyte
oAI = diffusivity of aluminium in the electrolyte
43
v = interfacial velocity between electrolyte and metal pad
J.l = viscosity of electrolyte
This then gives
kAI - O.023{~~ }Re083.sc033
= 0.02044. D AI . £-0.17 . p-O.5 • pO.5 . VO. S3 (29)
Lillebuen and Mellerud[41] then calculated how the current efficiency

should vary with the concentration of dissolved alumina in the electrolyte.
In order to do this it was necessary to know how all parameters vary with
alumina concentration. Using the following relationships the graph In
figure 3.2 was obtained for current efficiency vs alumina concentration.
DAI seems relatively insensitive to changes in alumina concentration and

is assumed constant. Values taken from work by Vetyukov[29].
Haupin's relationship for AC02 is used as given above.
The solubility of CO 2 in cryolite melts was measured by Bratland[43] and is

given as
0.03027 - 0.0006985· wt
(30)
1- 0.08989· wt
where wt = wtOjo of alumina dissolved in electrolyte.
This expression is valid for alumina concentrations between 1 and 8 wt°jo.
The solubility of aluminium has been measured by Vetyukov[44] and is

given as a weight percentage
C:I - -0.2877+0.0268·R+0.0003· T -0.0019·CaF2 (31)

- 0.0043 . ( wt - 12.5)
where R = wt ratio of NaF to AIF3 in electrolyte
44
95
Temperature 965°C
Weight Ratio 1.35
if94 ~ \
Calcium Fluoride 5%
-
>
()
-
ACO 4cm
c: 93
.-
CD
.-
()
=
w
C 92
...CD...
::::J
0
91
90
1 2 3 4 5 6 7 8
Alumina Concentration (wt%)
Figure 3.2 Cell Current Efficiency vs Alumina Concentration

T = temperature in degrees Celsius
CaF 2 = wt% CaF 2 in electrolyte
The calculated current efficiency shows a minimum at 4% AI 2 0 3 and the

presence of a minimum is borne out by many studies on laboratory
cells[40,45,46,47,48,49,5o,51,52,53,54]. The literature data is summarised in table
3.2. No data seems to be available for the variation of current efficiency
with alumina concentration in modern computer controlled point feed cells.
General operating conditions indicate that these cells need to operate at
low alumina concentrations for maximum current efficiency and do seem
able to attain higher current efficiencies than side break cells at higher
alumina concentrations. Comparison of the model with data obtained from
side break cells is difficult as it is virtually impossible to separate the effect
of alumina concentration from the effect of temperature which rises with
decreasing alumina concentration.
The model assumes a perfectly uniform distribution of alumina in the bath.

In practise oxygen ions are consumed at the anode and therefore alumina
concentration will be low in the vicinity of the gas bubbles. At the metal
pad the alumina concentration may be higher than in the bulk due to the
ability of alumina-bath agglomerates to float upon the pad surface. The
bath film near the metal is enriched with sodium fluoride increasing the
local bath ratio and hence the alumina solubility. These effects will reduce
the metal solubility at the cathode and the gas bubble area at the anodes.
This will decrease the dissolution rates of both species and cause a shift
in the minimum of the current efficiency curve.
Although this model shows reasonable results for the variation of current
efficiency with alumina concentration it does not include a sufficiently
accurate hydrodynamic model to show the effect of interelectrode gap.
Current efficiency has been measured for differing anode to cathode
distances in industrial cells by Rolseth et aH55] and at Anglesey Aluminium.
Both have found that the current efficiency varies linearly with ACO at
values above 3-4 cm but drops off rapidly below this. Lillebuen's model for
the dissolution of aluminium shows the rate of dissolution and hence the
current efficiency to be virtually independent of the anode cathode
distance. It may be that there are two mechanisms responsible for current
efficiency dependent upon the distance between the electrodes. At low
45
Reference Type of Cell Correlation Range
Gjerstad[46] Laboratory Cell Linearly increasing 0.5 - 12%
Belyaev[47] Laboratory Cell Minimum at 5% 1-11%
G rjotheim[48] Laboratory Cell Minimum at 4.5% 0-11%
Fi rsanova[49] Laboratory Cell Minimum at g% 4 - 10.5%
Bersimenko[5O] Laboratory Cell Linearly increasing 3 - 12%
Szeker[51] Laboratory Cell Minimum at 4.5% 2 - 12%
Abramov[52] Laboratory Cell Linearly increasing 5 - 12%
Schmitt[53] Side-break Cell Increasing 2 - 7%
Thonstad[54] Side-break Cell Increasing 2 - 7%
Poole[55] Side-break Cell Increasing 2 - 7%
Lillebuen[4O] Side-break Cell Increasing or const 0.5 - 6%
Table 3.2 Current Efficiency vs Alumina Concentration: Literature Data

ACOs it is possible that the reaction is between carbon dioxide gas and
aluminium or that waves in the metal pad come in contact with the anode
and cause loss of current efficiency due to shorting.
The correlation for the dissolution rate constant is unsatisfactory for a

number of reasons.
Firstly it is strictly applicable to mass transfer at the solid/liquid interface

for the case of fluid flow through a tube which has little in common with
the three-phase circulating fluid system of an aluminium cell. It also seems
unlikely that the definition of Reynolds number using twice the ACO as the
critical dimension is capable of describing the hydrodynamics of the cell.
Secondly it is only appropriate to turbulent flow. Employing values of

electrolyte density of 2050 kg m-3 , viscosity of 3x10-3 Ns m-2 , anode
cathode distance of 0.04 m and interfacial velocity of 0.10 m S-1 (Lillebuen
uses a value of 0.06) the Reynolds number is of the order 5,000 which
falls well short of the figure of 10,000 usually quoted for the onset of
turbulence in pipe flow.
Thirdly, as with previous models, no account is taken of the turbulent

effect of the gas bubbles. As a term the interfacial velocity is undesirable
as it is extremely difficult to measure in industrial cells.
The model would therefore be improved by the use of a mass transfer

correlation determined for a system more closely analogous both to the
geometry and hydrodynamics of an aluminium cell.
3.5.6 ModeJ of Vasiliadis

Vasiliadis[56] undertook the determination of an improved mass transfer
correlation by investigation of the system Hexan-1-01 and Water. This
system had the advantage of being observed at room temperature and
could be analysed by refractive index measurements. The aqueous
phase, analogous to aluminium, was rotated by a disc stirrer and the
evolution of carbon dioxide simulated by introducing air or nitrogen into
the hexan-1-01 / electrolyte phase through a porous plate. The liquid
phases were contained in a cylindrical vessel.
46
The experiments resulted in the following correlation
where Vf = velocity of anode gases

N = stirrer speed revs/s
suffix W refers to water
The volumetric flowrate of anode gases may be determined from the rate
of electrolysis at the anode. The number of moles of gas formed
1 10
(33)
2 Zo ·F
where = partial current

z = charge number
suffix 0 refers to electrolysis of oxide ions
F refers to electrolysis of fluoride ions
The electrolysis of oxide and fluoride are the only reactions at the anode
so
where = total current
Zo = 2 and zF =1 so the equation reduces to
I
- (35)
4·F
Assuming that the gases behave ideally then according to the universal
gas law
R·T
- n .-- (36)
9 P
47
where v9 =volumetric flowrate m 3 S-1
R = universal gas constant J mol- 1 K-1
T = temperature of gas oK
P = Pressure of gas N m- 2
The velocity of the anode gases is the volumetric flowrate for unit surface
area of anode
- -.- R·T
-.-1 i·R·T
- (37)
4·F P AA 4·F·P
where AA = Total anode area m2

= anodic current density A m-2
The speed of the stirrer has no direct counterpart in the cell but the
velocity at its tip may be equated to the maximum velocities found in the
metal pad.
Inserting the relationships and the values for the physical properties of
water into the equation gives
kAI - 2.147x 10- 11 .

(s-
. TJ1.42 .
(1 + 313.3· V 2 .5 ).
( ~ )0.33 . D~,67 (38)
where v = maximum velocity in the metal pad
This model is more attractive as it will directly relate current efficiency to

parameters that are known to affect it in practice. It also has the
advantage that it contains only parameters that are documented or may
be relatively easily measured. Unfortunately, no correlation was made
relating the depth of the metal pad to the dissolution rate constant
although the metal pad velocity should be a function of metal pad depth.
The experiment was also not strictly geometrically analogous as a
cylindrical rather than rectangular geometry was used.
48
3.5.7 Model of Haupin
At the Trondheim course Haupin[30] presented the following model for

mixed rate control. The model uses correlations formulated in both
Lillebuen's and Robl's models. No assumption is made as to which step is
controlling although the reaction between species is assumed to be
instantaneous. As in Lillebuen's model this requires that for CAl > 0,
Cco 2=0. The reaction is assumed to take place at the carbon dioxide
boundary layer and the concentration profile is shown in figure 3.3.
The rate of dissolution of metal through the metal boundary layer is given
rAldiss = AA,· kA' . (C~I - C AI(1)) (39)
where rAldiss = dissolution rate for aluminium

CAI (1) = concentration of aluminium at electrolyte/ aluminium
boundary.
The rate of convective transport of metal from the metal to carbon dioxide
boundary layers is given by
(40)
where rAlconv = rate of convective transport of aluminium

A conv - interfacial area for aluminium/electrolyte in
convective zone
convective rate constant
concentration of aluminium at carbon dioxide
boundary layer
The dissolution of carbon dioxide is enhanced by the reaction that is

assumed to take place at the carbon dioxide boundary layer. The rate of
dissolution is given by
(41 )
where = rate of dissolution of carbon dioxide
49
Dissolved CO2
ANODE
Dissolved AI
C=o
.... .... ....

....
.... ....
L...
c / concentration
CATHODE
Figure 3.3 Aluminium Concentration across CO2 Boundary Layer

Haupin uses the correlation presented by Lillebuen for the aluminium
dissolution rate constant and the correlation presented by Robl for the
convective rate constant. The rate constant for the dissolution of carbon
dioxide is given as
Deo2 _
--=- 1. 703
__ - 0.7630
_ _ _In _Reeo
_--=-2 Re o.s Sco.s (42)
Deo 2 1- 0.6272 In Re . cO 2.
eo2
where Dco2 = diffusivity of carbon dioxide in electrolyte

bco2 = carbon dioxide bubble diameter
Reco2 = Reynolds number for carbon dioxide boundary layer
_ ve . Pe . Deo 2
Ile
Ve = velocity of the electrolyte

Pe = density of electrolyte
Ile = viscosity of electrolyte
Although this model makes no assumptions of the rate determining step it

is subject to the weaknesses in the correlations for the rate constants
already mentioned in the previous models. Although a correlation for the
carbon dioxide dissolution rate constant is presented, its use is limited by
the lack of any correlation for bubble diameter and electrolyte properties
and the difficulty in obtaining a value for the velocity of electrolyte within
an industrial cell. It does however, give an indication of some of the
parameters likely to be of importance in the improvement of current
efficiency.
By requiring the concentrations of aluminium at both boundary layers the

uniform distribution of alumina throughout the bath may not be assumed.
Although this will lead to a more accurate model, its use is not so
desirable in a dynamic simulation as it increases the number of equations
required and the complexity of the solution. If the electrolyte may be
assumed to be turbulent then the rate of convective transport may be very
much greater than the dissolution rate and so the model will approximate
that presented by Lillebuen.
50
3.6 Models for the Estimation of Frozen Bath Shape
Aluminium reduction cells are designed such that a layer of frozen bath
forms around the sidewalls of the cell. This protects the material of the
sidewalls from the corrosive effects of the electrolyte and the metal pad.
The frozen layer also helps to regulate the temperature of the cell by
freezing and melting. The freeze profile also affects the flow patterns
within the cell and the current distribution.
Models for the precise estimation of freeze shapes have been developed
by a number of workers[57,58,59]. These involve the division of the cell into
a number of triangular elements for which the temperature distribution can
be calculated by a finite difference method. The accuracy of the simulation
depends upon the number of lattice points used. This type of model is
therefore unsuitable for dynamic simulation due to the time required for
these calculations.
A simpler model for heat transfer through the side walls assumes
unidirectional heat flow through the walls. This model has been used by a
number of workers for the calculation of steady state freeze
thicknesses[60,61] and for the dynamic behaviour of freezes[2,5,62]. The heat
flow and temperature gradients are shown schematically in figure 3.4.
The fundamental equations for the flow of heat through the side walls are
qm - Am . hm. (t m- t 1 ) (43)
qn - An .~.(tn -t n+1) (44)

Xn
qs - As . hs . (t s - ta ) (45)
where qj = heat flow from section i

h·I = heat transfer coefficient from section i
k·I = thermal conductivity of section i
X,I = thickness of section i
t·I = temperature at surface of section i or in the case of the
melt and the surroundings, the bulk temperatures
A·I = area for heat transfer from section i
51
electrolyte
lj
/
freeze
~
carbon
sidewall
•
firebrick
t
steel
casing
" /
Aluminium
pad
Tb
Figure 3.4 Heat Flow and Temperature Gradient across Frozen Sidewalls
and suffix
m represents the melt ( bath or metal)
n represents a section of the wall or ledge with n=1 being
the freeze and n=m being the outer shell
s represents the outer shell
a represents the cell surroundings
The heat transfer at the inner and outer surfaces is assumed to be purely
by convection.
An imbalance in the energy conducted into and out of the ledge/melt

boundary will cause freezing or melting of the ledge. The heat imbalance
will be made up by the latent heat associated with this melting or freezing.
A heat balance at the freeze surface gives
qm - Am ·h m·(t m-t 1 )
- A,· ~: . (t, - U + (A + Cp· (t m - I,)). ~~ (46)
where A = latent heat of fusion of electrolyte

dm
dt = rate of change of mass of freeze
Cp = specific heat capacity of boundary layer
Most models have assumed that the temperature at the bath ledge
interface is the liquidus temperature of the bath. This is a reasonable
assumption if the melt can be assumed to be well mixed and the diffusion
of components away from the boundary layer fast.
If the electrolyte behaves as a simple eutectic system then the frozen

ledge will consist of one pure solid compound or solid solution. If,
however, freezing is much faster than the diffusion of species from the
boundary layer. Chemical analysis of ledge in industrial cells by
Thonstad[61] has shown it to consist of rather pure cryolite with smaller
quantities of alumina and calcium fluoride than were present in the bath.
Samples taken from the metal and bath ledges in a number of 'dead ' cells
at Anglesey had a weight ratio of 1.36-1.41 irrespective of the ratio in the
52
bath. This indicates that there is diffusion of individual components from
the boundary layer to the melt.
This effect has been modelled by Hashimoto[51. The diffusion of species i

is given by
aC j _ 0.. a2 c j u. aC j
at I ax 2
ax (47)
where OJ is the diffusion coefficient of species i and U is the velocity of

solidification and is assumed constant. This equation was solved by
applying the following approximations.
(a) The mass of the components moving between the freeze and the
boundary layer is calculated from the amount of the electrolyte which
freezes or melts.
(b) The concentrations of the components are represented by their

average values in the boundary layer.
The mass of the components moving between the bulk electrolyte and the
boundary layer is then calculated
(48)
where Wj = the amount of species i entering the boundary layer from

the bulk
W f = the amount of bath in the boundary layer per unit area
f = a fittable constant related to the diffusivity
C j,f = average concentration of species i in the boundary layer
C j ,I = average concentration of species i in the bulk
The boundary layer is bath that is freezing or melting and W f is calculated

from equation (48). The value of f is estimated by Hashimoto from
measurements in commercial cells and ranges between 0.1 and 0.00001.
The exchange of mass between the layer and the bulk is taken to be
relatively rapid for the layer in the bath but slow for the metal zone. The
value of f has a marked effect upon the freeze thickness as is shown in
figure 3.5. In this case it appears that the diffusivity of the specific
53
20
- E
(,)
t/)
t/)
C1l
c:
18
~
.~
.c:
I-
C1l
N
C1l 16
...
C1l
11. f = 0.00001
14
o 1 2 3 4 5 6
1000
/ Bath temperature
--
o
...C1l
:::::J
~
980
Melting point of bath

C1l
Co
E
{!!. 960
940
o 1 2 3 4 5 6
Time I hr
Figure 3.5 Variation of Freeze Thickness with Parameter f

components have been used to limit the amount of freeze formed by
equation (48). In fact, the imbalance of heat flow at the boundary layer
must be provided as latent heat by freezing or melting of the layer.
Equation (47) cannot limit the total mass of the boundary layer that
freezes or melts, but limits the composition that this layer can attain.
It can be seen from figure 3.5 that as f is reduced the freeze becomes
more stable to changes in temperature. If it is to be assumed that f is very
little for the metal zone then the freeze profile in this zone will be stable.
Steady state models have found the ledge in the metal zone to be at least
as responsive to temperature as that in the bath zone if not more so. The
assumption that the diffusivity is less in the metal zone is also not borne
out by the measurements at Anglesey. These found the same ledge
composition in both zones.
3.6.1 Determination of heat transfer coefficients
A number of workers have calculated heat transfer coefficients for melts to

sidewalls based upon correlations between dimensionless groups, the
correlations being dependent upon the geometry of the system. The
commonly used dimensionless groups are shown in table 3.3.
Heat transfer coefficients will vary throughout the cell dependent upon
local conditions. An approximate heat transfer coefficient may be obtained
either by calculation using average conditions or by averaging the values
obtained for expected extremes of conditions.
3.6.2 Model of Haupiri-60]
Bennet and Meyers[63] give the following correlation for the heat transfer
coefficient of fluids, other than metals, flowing over a plate at Re < 5 x 105 .
Nu = 0.664 Pr1/3 Re 1/2 (49)
where the physical dimension for evaluating the Nusselt and Reynolds
numbers is the interface dimension parallel to the fluid flow.
For aluminium flowing through a pipe Filimonov et aH64] gives the following
correlation
54
Number Symbol Equation
h·L
Nusselt Nu -
k
L3. p3 .g.~T
Grashof Gr 112
Cp'l1
Prandtl Pr
k
Rayleigh Ra Gr·Pr
L·v·p
Reynolds Re
11
where h = heat transfer coefficient of fluid

L = physical dimension for description of system
diameter in pipes of circular cross-section
k = thermal conductivity of fluid
p = density of fluid
g = acceleration due to gravity
~T = temperature gradient across fluid
11 = viscosity of fluid
Cp = specific heat capacity of fluid
v = velocity of the fluid
Table 3.3 Commonly Used Dimensionless Groups for Determination of

Heat Transfer Coefficients
Nu = 5 + 0.025 ( Pr·Re )0.8 (50)
Using these correlations, Haupin calculated the heat transfer coefficients

for the metal pad to ledge and the bath to ledge. The critical dimension for
the electrolyte was taken to be the bath depth measured at an angle to
account for electromagnetic stirring and the drag force of swirling metal.
The equivalent diameter for fluid flowing in a shallow stream is used in the
correlation for the metal. This is given by
4·H m ·R m
L = (51 )
where Hm = the depth of the metal pad

Rm = is the radius of rotation of molten metal
Applying typical properties for the metal and electrolyte at 980°C Haupin
obtained values of hb = 370 W m- 2 K-1
and hm = 1500 W m- 2 K-1. The
result for hb compares well with values calculated from measured heat
fluxes but the calculated heat flux in the metal zone was 30-40% high.
This discrepancy was attributed to the additional thermal resistance of the
film of molten electrolyte always present at the metal/ledge interface.
The effective heat transfer coefficient is then given by
1
(52)
where h me = effective heat transfer coefficient in metal zone

hm = calculated heat transfer coefficient
xe = thickness of electrolyte film
ke = thermal conductivity of electrolyte
Assumption of a film thickness of 0.03 cm resulted in an effective heat

transfer coefficient of 1100 W m-2 K-1.
3.6.3 Model of A.Solheim and J. ThonstacJ.65)

Solheim and Thonstad used a low temperature model cell with gas
induced flow to study the effect of gas flowrate, anode immersion depth
55
and anode-ledge distance on the heat transfer coefficient between the
bath and ledge. This resulted in the relationship
GF .AIJk2
hb(gas) - k1· ( (53)
ALD
where hb(gas) = heat transfer coefficient due to gas induced flow

only
GF = gas flow rate per unit length of anode periphery
AI = anode immersion depth
ALD = anode-ledge distance
k1 ,k2 depend upon the position down the sidewall. For the
middle position k1 = 140, k2 = 0.42.
This gives heat transfer coefficients in the region of 200-500 W m- 2 K-1 but
does not include the rotational effect of the bath due to electromagnetic
forces. By applying a correction factor to account for horizontal flow the
equation becomes
(54)
where hb = heat transfer coefficient from bath to ledge

vh = horizontal component of resultant fluid velocity
Vv = vertical component of resultant fluid velocity
Increasing the horizontal component from 0 to Vv results in a 19% rise in

the value of hb giving heat transfer coefficients of between 250-
600 W m-2 K-1 in the middle region of the cell. This compares well with
other published data.
3.6.4 Model of B.J. Welch

Welch[66] assumes a Nusselt correlation for natural convection
superimposed with forced convection. Experimental data for heat flux
measurements on immersed rotatable graphite cylinders giving flow
patterns similar to those expected in industrial cells shows a straight line
relationship between the Nusselt and Rayleigh numbers. The data for both
56
natural convection and natural convection plus superimposed velocity of
approximately 0.15 m S-1 is shown in figure 3.6. Whilst it probably
represents the most reliable data published, the accuracies for the heat
transfer coefficients is only +20%.
For natural convection involving parallel plates the form of correlation is
Nu = a·(Gr·Pr)m = a·Ra m (55)
where a and m vary with orientation and size. Typically a varies between
0.2 and 1.3 and m is usually between 0 and 0.33.
For forced convection with laminar flow the correlation is usually of the
form
(56)
depending upon the geometry, a varies between 0.1 and 1 m between

0.33 and 0.8 and n about 0.33.
Using these correlations and parameters obtained from the experimental

data Welch estimates ranges of heat transfer coefficient throughout the
cell. For a bath with 10°C superheat, that is 10°C above the liquidus
temperature of the bath, the local heat transfer coefficient for situations
where only natural convection occurs is calculated as
450 < hb < 480 W m- 2 K-1
For the combined model with fluid velocities up to 0.15 m S-1 the range in
heat transfer coefficient is
650 < hb < 830 W m-2 K-1
This data shows that the heat transfer coefficient can vary by a factor of
two for various locations on a cell IS sidewall. Although these values are
higher than other estimates published in literature Welch claims them to
have been substantiated by independent unpublished data.
57
1000
o Data for Run 1 x < 40mm
• Data for Run 4 60 RPM
, ...
- •• - •• Regression Line for Run 1
::l
Z 100 Regression Line for Run 4
Natural plus superimposed
forced convection
, .
~
..-
.. ... .
ci
-
z
a;
In
In
- - - - Flat Plate Correlation for Free
Convection
__ ----_e- -----_. - -.
• _.-D---a--- ---
..
....... . .
::l
Z
cau _.'
__ - - . -
0----- . . []- -- .0 _ - _ 0-
o
..J 10 •- ---
Natural convection
1
1.00E+04 1.00E+05 1.00E+06 1.00E+07 1.00E+08
Local Rayleigh No. Ra
Figure 3.6 Local Nusselt Numbers as a Function of Local Rayleigh Number

This model does not include a specific correlation for gas induced
turbulence in the cells. This is accounted for by the choices of the
coefficients used in the two correlations, the velocity and critical dimension
used in the calculation of the Reynolds number and degree to which the
two models are superimposed.
3.7 Models for the Voltage Drop across the Cell
The voltage drop across the cell may be broadly split into external and
internal voltages. The external voltages are those external to the
electrolyte and metal pad and include the ohmic drops across the buswork
of the cell, the anodes and the cathode. The internal voltage drops include
the ohmic drops across the electrolyte and the metal pad and the voltage
required for the electrochemical reactions in the bath.
The cell voltage can be broken down into components
(57)
where E0 = thermodynamic equilibrium potential for the reaction

AI 20 3 + 1.5C ~ 2AI + 1.5C02
llA = overpotential at the anode
llc = overpotential at the cathode
I = cell cu rrent
RA = electrical resistance of anode assembly
Rs = electrical resistance of bath
RM = electrical resistance of metal pad
Rc = electrical resistance of cathode assembly
Rx = electrical resistance of buswork external to the cell
The electrical resistance for the anode assembly, cathode assembly and
external buswork can all be simply measured on an operating cell.
Although the ohmic drop across the external buswork does not affect the
heat balance on the cell, it must be included to determine the power
efficiency of the cell.
The general equation for the resistance for a homogenous material with
uniform temperature is
58
1(.£
R - (58)
A
where R = resistance of object

1( = electrical conductivity of material
£ = length of object
A = cross-sectional area for current flow
The resistance of the bath and metal pad may be approximated by the
above relationship when an effective area for current flow is used. The
effective area is an approximation to the true current distribution of the
cell. Accurate models for the calculation of current distribution and voltage
profiles are too complex for use in dynamic simulation. In models where
the temperature and composition of the electrolyte is assumed to be
uniform throughout, the current may be assumed to flow uniformly through
an effective area based upon the area of the electrode. Haupin[67]
correlated the effective anode cross sectional area with anode
environment by determining the isopotential planes under and along the
sides of each anode in a laboratory cell. This was done using an
aluminium coated tungsten reference electrode and a micrometer
positioning device. The effective area was correlated to anode enviroment
by fanning factors, defined as an additional length or width added to the
actual dimensions of the anode. The effective area of an anode is then
(59)
where Ai = effective area for current flow under anode i

LA = actual length of anode
W A = actual width of anode
FL l' FL2 = fanning factors in the two directions of length
Fw1 , FW2 = fanning factors in the two directions of width
A typical set of fanning factors determined against anode-cathode

distance is shown in figure 3.7.
Haupin assumes that the effective area for the bath is equal to the total
effective area of the bank of n anodes and is given by
59
12
2 Facing another anode

7.6 cm spacing or an
insulating wall at 3.8 cm
o
o 2 4 6 8
Anode-cathode distance (cm)
Figure 3.7 Fanning Factors for Determination of Effective

Anode Area
n
AA - AB - LAi (60)
i=1
where AA = effective area for current flow through bank of anodes

As = effective area for current flow through bath
n = number of anodes in cell
This value is used to calculate the current density at the anode and the
electrical resistance of the electrolyte.
Haupin[30] considers the resistance of the bath to be composed of two

parts, the resistance of the electrolyte, Re , and the resistance due to
bubbles adhering to the anodes, Rba . The resistance of the bath is then
(61 )
Haupin quotes the electrolyte resistance as
d-8 (62)
AB .1(
where d = anode-cathode distance

ta = thickness of monolayer of bubbles ahering to anode
1( = electrolyte conductivity uncorrected for bubbles
8 = total thickness of bubble layer
E = gas fraction in free bubble layer
and the function (1 - E ) 1.5 comes from Bruggeman[68]
The resistance of bubbles adhering to the anode derived from data

reported by Sides and Tobias[69] is
(63)
- 1( . ( 1- 1. 26fJ . A A
where fc = fraction of the anode surface covered with bubbles
Haupin quotes this equation valid up to fc = 0.65. This model for the
electrolyte resistance is shown schematically in figure 3.8.
60
Rs - Re + Rba
Re - d - 0 o - ta
+ As K (1_£)1.5
As K
Rba - til
K (1 - 1.26 fc) AA
Anode
tA
000
d
Electrolyte
Aluminium Pad
1
Figure 3.8 Schematic Diagram of the Model for Resistance across the Bath
The accuracy of this model is dependent upon the relationships used for
fc' £, ta and 8. In calculations at the Trondheim course the following
relationships for fc and £ are used and ta and 8 are assumed constant.
100
0/0 of anode surface (64)
where 0/oA1 2 0 3 is the weight percentage of AI 2 0 3 in solution in the

electrolyte
No indication is given of how these relationships were derived or the type

of cell that they were derived for. The relationships are only related to the
alumina concentration in the bath but both properties will be affected by
the composition and temperature of the bath.
3.7. 1 Electrical conductivity

The electrical conductivity of the electrolyte is a function of the
composition and temperature of the electrolyte. There have been
numerous experimental measurements of the electrolyte and regression
analysis by Choudhary[70] led to the following equation
InK = 2.0156- 2068.4 +0.4349(Ratio)-0.0207(0/oAI 2 0 3)

T
- 0.005(0/oCaF2 ) - 0.0166(0/oMgF2 ) + 0.0178(0/oLiF)
+ O. 0077( 0/oLi3AIF6 ) + O. 0063( %NaCI) (65)
where k = electrical conductivity ,Q-1 cm- 1

T = temperature K
Ratio = weight ratio NaF/AIF3
% refers to weight percentage of component
This equation is claimed to give values of electrical conductivity with an

accuracy better than + 4%.
Three theoretical models have been presented by P. Fellner et a1[711. The

first model is based on the assumption that the conductivity is proportional
61
to the number density of the effective electrical charges in the melt. In
cryolite melts with large complex anions the current is carried mainly by
cations which may be assumed to move by the influence of an electric
field relative to the framework formed by the anions.
As a first approximation to the conductivity Fellner obtained
1\:mix (66)
where k mix = conductivity of melt

xi = mol fraction of component i in the melt
zi = effective charge of the current carrying species
ViI/> = molar volume of species i
The numerical value of 93.5 Q-1 cm 2 mol- 1 for the proportionality factor
was determined experimentally from the conductivity of pure cryolite by
extrapolation to its value at 1273 K. This factor is assumed constant when
the anion framework is not significantly affected by the additions.
The two other models assume that the Charge-carrying species do not
mutually influence each other. The conductivity is then the sum of the
contributions of the pure melts placed in two conductivity cells connected
in series or parallel. A schematic diagram of these two models is shown in
figure 3.9. For two melts with conductivities 1\:1 and 1\:2 and volumes V 1 and
V 2 in conductivity cells with cross sectional area A the resistivities R1 and
R2 are given
(67)
In the series model, the conductivity cells are connected in series and the
total resistivity, Rtotal is the sum of R1 and R2
+ (68)
where "-mix,s denotes the conductivity of the system for the two cells
1;""
connected in series and is equal to the ideal conductivity of the mixture of
62
Series model Parallel model
Schematic illustration of models for the calculation

of conductivity of a binary mixture with volume
fractions of the ratio 1:2.
Figure 3.9 Models for Electrolyte Conductivity

the two melts. Introducing molar volumes and mole fractions and
extending the model to mUlti-component mixtures gives
~x.I ·v~I
£.
K:mix,s (69)
Fellner states that this model can be shown to give a good description of
systems that exhibit nearly ideal thermodynamic behaviour.
The parallel model, following the procedure for the series model, gives the
following expression for the electrical conductivity "mix,p
~ x.I 'K:.I ·v~I

£.
K:mix,p (70)
This model is a boundary model and the electrical conductivity is always

less than that calculated.
The calculated data using all three of the models show good agreement
with experimentally measured values for the electrical conductivity at
various bath compositions used industrially. As shown in figure 3.10 the
simple model shows very good agreement for binary cryolite melts at
1273 K with the exception of the Na3AIF6 - LiF system containing more
than 5 wtO/o LiF. The experimental data for this system falls between the
predicted values of the series and parallel models approximating to the
former at low wtO/o LiF and to the latter at high wtO/o LiF.
Data calculated from the simple model for the tertiary system Na3AIF6 -
AI 2 0 3 - LiF is shown in figure 3.11. In this system AI 2 0 3 decreases the
conductivity while LiF increases it. This then provides a good test for the
suitability of the simple model to predict conductivity data. The data shows
good agreement with measurements by Matiasovsky et aH72 1.
The electrical conductivity of molten aluminium is such that the resistance

of the metal pad may be neglected or assumed constant and included in
the overall resistance of the cathode.
63
3.4
3.2
3
,....
I:(,) 2.8
a:c..
2.6 MgF2 ----
2.4
2.2 AI203
2
0 5 10 15 20 25 30 35 40 45 50
Na3A1Fs + Additions, mol%
Figure 3. 10 Results of Fellners Simple Model for Binary Systems

3.2
3 x o rrolO/o AI203 u
--
0
0.....;--
,....
Ie
2.8 F
x 10 rrol% AI203
0 -
(,) 0
a "II
2.6 u
~
0
x = 20 rrol% AI203
2.4 0
2.2
2 I I
o 10 20 30 40
NCia AIFs + x AI 2 0 3 mol % LiF
Figure 3.11 Results of Fellners Simple Model of Tertiary System

3.7.2 Reversible decomposition potential and electrode overpotential
For an overall cell reaction there is a thermodynamically reversible

decomposition potential defined as
~G¢'
(71 )
z·F
where ~G ¢' = Gibbs free energy for the reaction

z = number of electrons transferred in the reaction
F = Faraday's Constant
For a non ideal system this becomes
R· T ( rra~ )
+ --In (72)
z·F
where aj = activity of reagent i

v = stoich iometric coefficient for reagent i
There is also some excess voltage above the reversible range required for
the overall reaction. This is the reaction overpotential and may be
considered to consist of an anodic part and a cathodic part.
The overall reaction for aluminium electrolysis under normal operating

conditions is generally assumed to be
[4]
The ideal thermodynamic decomposition voltage for this reaction IS
approximately 1.2 V[73] and is relatively unaffected by variations in

temperature and electrolyte compostion within normal operating
conditions. If the system is considered to be non-ideal then the
decomposition voltage becomes
(73)
64
Haupin[74] assumes the activities of all reactants and products, except
alumina, to be unity. This is valid if these components may be assumed to
exist in their standard states, that is that the aluminium and carbon are
essentially pure condensed phases and that nearly pure carbon dioxide
contacts the anode at approximately 1 atm pressure. Alumina in solution
will have unit activity only at saturation and Haupin uses the following
relationship for alumina activity
(74)
where CAI203 = alumina concentration wtO/o

Csat = saturation concentration of alumina wt%
This relationship is derived from work by Dewing[75] on alumina activity in

pure cryolite at 1300 K.
Under normal electrolysis and non-equilibrium conditions, the overall cell

reaction voltage is higher than the reversible decomposition potential. This
is because excess voltage above the reversible value is required for the
reaction to proceed at practical rates. The voltage in excess of the
reversible decomposition potential is the cell overpotential.
The total overpotential across the cell consists of an overpotential at each

electrode and investigations of cell potential generally consider the
cathodic and anodic overpotentials separately. In the aluminium reduction
cell, the overpotential is primarily affected by alumina concentration,
anode carbon characteristics, current density and temperature.
Haupin[30] considers the overpotential at an electrode to be composed of

two parts. These are a surface or reaction overpotential, 11s' and a
concentration overpotential, 11e' The total overpotential at the electrode, 11t,
is then
11t = 11e + 11s (75)
65
Under different circumstances the concentration and reaction
overpotentials have greatly different values and the reaction can become
concentration or surface limited
3.7.3 Cathodic overpotentiai
Many investigators have measured cathodic overpotential and found it

fitted a Tafel relationship over the range of commercial interest.
Laboratory measurements by Thonstad and Rolseth[76], for electrolytes of
industrial interest without stirring, gave Tafel plots
1] =a+ f3 log i (76)
where 11 = electrode overpotential V

= electrode current density for reaction A cm- 2
a, f3= Tafel coefficients
with coefficient a ranging from 0.19 to 0.21 and f3 ranging from 0.23 to
0.25. Stirring lowered the overpotential by 40-50% which Haupin[30] takes
to indicate a concentration overvoltage. Thonstad and Rolseth found the
cathodic overpotential in industrial cells to be between the values for
unstirred and vigorously stirred laboratory cells at about 0.1 V. Alcoa data
for industrial cells[77] resulted in the following empirical relationship valid
for current densities above 0.1 A cm- 2 .
R· T· (1.375 - 0.125· CR) In ( ic J (77)

1.5· F 0.257
where 1]et = total overpotential at the cathode

CR = cryolite ratio, mole ratio NaF/AIF3
Ie = cathodic current density
66
References
1 Richard, M. C., "Use of a Mathematical Cell Model to Determine Cell

Parameter Design Changes for Production Maximization", AIME Light
Metals, 1975, pp95
2 Sulmont, B. and Hudault, G., "Application of a Thermoelectric Model to

the Investigation of Reduction Cell Thermal Equilibrium", AIME Light
Metals, 1978, pp73
3 Blake, S. R., "A Model for the Variation of Alumina Concentration in

Industrial Hall-Heroult Cells", MSc Thesis, Department of Chemical
Engineering, University of Newcastle upon Tyne, June 1983
4 Ek, A. and Fladmark, G. E., "Simulation of Thermal, Electric and

Chemical Behaviour of an Aluminum Cell on a Digital Computer", AIME
Light Metals, 1973, pp85
5 Hashimoto, T. and Ikeuchi, H., "Computer Simulation of Dynamic

Behaviour of an Aluminum Reduction Cell", AIME Light Metals, 1980,
pp273
6 Thonstad, J., Nordmo, F. and Paulsen, J. B., "Dissolution of Alumina in

Molten Cryolite", Metallurgical Transactions, vol 3, February 1972,
pp403-407
7 Gerlach, J., Hennig, U. and Kern, K., "The Dissolution of Aluminium

Oxide in Cryolite Melts", AIME Light Metals, vol 1, 1974, pp49-61
8 Thonstad, J., "The 4th Int. Course on Process Metallurgy of

Aluminium", Trondheim, 1985
9 Winkhaus, G., "On the Dissolution Rate of Alumina in Cryolite Melts",

99th AIME Annual Meeting, Denver,1970, Paper A70-25
10 Maeda, H., Matsui, S., and Era, A., "Measurement of Dissolution Rate
of Alumina in Cryolite Melt", AIME Light Metals, 1985 pp763-780
67
11 Asbj0rnsen, O. A., Andersen, J. A., "Kinetics and Transport Processes
in the Dissolution of Aluminium Oxide in Cryolite Melts", AIME Light
Metals 1977, pp137-152
12 Kachanovskaya, I. S., Osivik, V. I. and Kukhotkina, T. N., "Phase

Transformations and Alumina Dissolution in Cryolite", Tsvetnye Metally,
No 12, 1971, pp10-23
13 Jain, R. K. et ai, "Interaction of Aluminas with Aluminium Smelting

Electrolytes", AIME Light Metals, 1983, pp609
14 Thonstad, J., Johansen, P., and Kristensen, E. W., "Some Properties of

Alumina Sludge", AIME Light Metals, 1980, pp227-241
15 Hove, S.J. and Kvande, H., "Centre-Break Alumina Feeding and Sludge
Control of Prebaked Cel/s", AIME Light Metals, 1982, pp513
16 Keller, R., "Alumina Dissolution and Sludge Formation", AIME Light

Metals, 1984, pp513
17 Less, L. N., "The Crusting Behaviour of Smelter Aluminas", AIME Light

Metals, 1976, pp315
18 Thonstad, J., R0nning, S. and Entner, P., "Formation of Bottom Crusts

in Aluminium Pots. A Laboratory Study", AIME Light Metals, 1982,
pp485
19 Bagshaw, A. N., Kuschel, G., Taylor, M. P., Tricklebank, S. B., and

Welch, B. J., "Effect of Operating Conditions on the Dissolution of
Primary and Secondary (Reacted) Alumina Powders in Electrolytes",
AIME Light Metals, 1985, pp649-660
20 Entner, P, Scmidt-Hatting, W., Mitrovic, Z., Gruber, U., Rufer, D. and

Fj0rnes, "Investigation of the Dynamic Behaviour of Aluminum Pots",
AIME Light Metals, 1984, pp701
68
21 Haupin, W. E., "Mathematical Model of Fluoride Evolution from Hall-
Heroult Cells", The 4th Int. Course on Process Metallurgy of Aluminium,
Trondheim, 1985
22 Sidorov, L.N., Kolosov, E.N. and Shol'ts, V.B., Russian Journal of

Physical Chemistry, 42, pp1382-1386, 1968
23 Grjotheim, K., Kvande, H. and Motzfeldt, K., "Vapor-Liquid Equilibria in

the System NaF-AIF3-AI203'" AIME Light Metals, 1975, pp125-137
24 Newitt, Dombrowski and Knellman, Trans. Inst. Chem. Engr., vol 32,
1954,p244
25 Azbel, D.S. and Lee, S.L., "Two-Phase Momentum Heat Transfer

Chemistry", Process Eng. Syst., vol 1, p159, 1978
26 Grjotheim, K., Haupin, W.E and Welch, B. J., "Current Efficiency--

Relating Fundamental Studies to Practise", AIME Light Metals, 1985,
pp679
27 Frovlova, E.B., Dobrokhotov, V.B. and Tsyplakov, A.M., Trudy V.A.M.I,

89, vol 36, 1971
28 Waddington, J. and Pearson, T. G., "Electrode Reactions in the

Aluminium Reduction Cell", Discussions, Faraday Society, 1947, vol 1,
pp307-320
29 Vetyukov M. M. et ai, Isv. Vyssh. Uchebn. Zaved., Tsvetn. Metall.,

1978,p62
30 Haupin, W. E., "The 4th Int. Course on Process Metallurgy of

Aluminium ", Trondheim, 1985
31 Qiu, Z. and Fan, L., "The Rate Determining Step of Metal Loss in
Cryolite-Alumina Melts", AIME Light Metals, 1984, pp789
69
32 Poole, R.T., Etheridge, C., "Aluminium Reduction Cell Variables and
Operations in Relation to Current Efficiency", AIME Light Metals, 1975,
pp163-183
33 Wittner, H., "Working Methods for High-Current Efficiency and Flexible

Amperage", AIME Light Metals, 1975, pp139-150
34 Wittner, H. and Giulini, Gebr., "How to run a Smelter with Current

Efficiency higher than 90%-Scientific and Practical Basis", AIME Light
Metals, 1977, pp153-161
35 Bratland, D., Jose del Campo, J. and Cho, K., "Current Efficiency
Measurements in Laboratory Aluminium Cells. Influence of Bath
Volume, Anode Size and Anode Quality", AIME Light Metals, 1981,
pp281-308
36 Berge, B., Grjotheim, K., Krohn, C., Naeumann, R. and T0rklep, K.,
"The Influence of Operating Parameters on the Current Efficiency in the
Aluminium Reduction Cells", AIME Light Metals, vol 1, 1976, pp23
37 Robl, R. F., Haupin, W. E., and Sharma, D., "Estimation of Current

Efficiency by a Mathematical Model Including Hydrodynamics
Parameters", AIME Light Metals, vol 1, 1977, p185
38 Evans, J.W., Zundelevich, Y. and Sharma, D., Met. Trans, vol 12B,
1981
39 Levich, V.G, "Physicochemical Hydrodynamics", Prentice Hall, New

Jersey, 1962, pp690-698
40 Lillebuen, B., Ytterdahl, S. A., Huglen, R., and Paulsen, K. A., "Current
Efficiency and Back Reaction in Aluminium Electrolysis", Electrochimica
Acta, vol 25, No, 2, February 1980, p131
41 Lillebuen, B., and Mellerud, Th., "Current Efficiency and Alumina

Concentration", AIME Light Metals, 1985, p637
70
42 Aarebrot, E. et ai, Metall. vol 32, no 41, 1978
43 Bratland, D., Grjotheim, K., Krohn, C. and Motzfeldt, K., Journal of

Metals, vol 19, 1967, pp13
44 Vetyukov, M. M. et ai, Physical Chem. and Electrochem. of Molten Salts

and Slags, Kiev, 1969, pp367
45 Gjerstad et ai, Symposium on Industrial Electrode Processes with Gas

Evolution, Electrochem. Soc., 1966
46 Belyaev et ai, Izv, Vyssh. Ucheb. Zav., Tsvetn. Met., vol 4, no 3, 1961,
p67
47 Grjotheim K. et ai, Can. Met. Quart., vol 11 , 1972, p295
48 Firsanova et ai, Izx. Vyssh. Ucheb. Zav., Tsvetn. Met., vol 5, no 3,

1962, p52
49 Bersimenko et ai, Tsvet. Met., vol 46, no 3, 1973, p30
50 Szeker, Acta Tech. Acad. Sci. Hung., vol 10, 1955, p91
51 Abramov et ai, "Theoretical Principles of the Electrometallurgy of

Aluminium", Moscow, 1953
52 Schmitt, H., "Extractive Metallurgy of Aluminium", vol2, Intersci. Pub.,

New York, 1963, p169
53 Thonstad, J. et ai, Proceedings, 3rd Czech. Aluminium Symposium,

Banska Bystrica, 1976, p88
54 Poole, R. T. and Etheridge, C., "Aluminium Reduction Cell Variables

and Operations in Relation to Current Efficiency", AIME Light Metals,
1977,pp163
71
55 Rolseth, S. and Thonstad, J., "On the Mechanism of the Reoxidation
Reaction in Aluminum Electrolysis", AIME Light Metals, 1981, pp289
56 Vasiliadis, C. G., "Mass Transfer Studies related to Aluminium

Electrolysis", MSc Dissertation, Oct 1981, University of Newcastle upon
Tyne
57 Arita, Y., Urata., N. and Ikeuchi, H., "Estimation of Frozen Bath Shape
in an Aluminum Reduction Cell by Computer Simulation", AIME Light
Metals, 1978, pp59

Metals, 1978, pp73
59 Peacey, J. G. and Medlin, G. W., "Cell Sidewall Studies at Noranda

Aluminum", AIME Light Metals, 1979, pp475
60 Haupin, W. E., "Calculating Thickness of Containing Walls Frozen from

Melt", AIME Light Metals, 1971, pp188
61 Thonstad, J. and Rolseth, S., "Equilibrium between Bath and Side

Ledge in Aluminium Cells. Basic Principles", AIME Light Metals, 1983,
pp415
62 Paulsen, K.A., Huglen, R., Andersen, J.A., Lillebuen, B., and Ytterdahl,
S.A., "Variation of Side Lining Temperature, Anode Position and
CurrenWoltage Load in Aluminium Reduction Cells", AIME Light
Metals, 1980, pp325-339
63 Bennet and Meyers, "Momentum, Heat and Mass Transfer", McGraw-

Hill, 1962
64 Filimonov, S. S., Kryukova, M. G. and Teplov, S. V., Thermal Physics of

High Temperatures, vol 2, no 6, 1964, pp901
72
65 Solheim, A. and Thonstad, J., "Heat Transfer Coefficients between Bath
and Side Ledge in Aluminum Cells. Model Experiments", AIME Light
Metals, 1983, pp425
66 Welch, B.J., "Energy Balance-Transient Thermal Effects During Cell

Operation", The 4th Int. Course on Process Metallurgy of Aluminium,
Trondheim, 1985
67 Haupin, W.E., "A Scanning Reference Electrode for Voltage Contours

in Aluminum Smelting Cells", Journal of Metals, Oct., 1971
68 Bruggeman, D.A., Ann. Physik, vol 24, 1935, p632
69 Sides, P.J. and Tobias, C.W., J. Electrochem. Soc., vol 129, 1982,
p2715
70 Choudhary, G., "Electrical Conductivity for Aluminum Cell Electrolyte

between 9Scr-102SOC by Regression Equation", J. Electrochem. Soc.,
vol 120, 1973, pp381-383
71 Fellner, P., Grjotheim, K. and Kvande, K., "Model Calculations of

Cryolite Melts", AIME Light Metals, 1984, pp805
72 Matiasovsky, K., Danek, V. and Malinovsky, M., "Effect of LiF and

Li3AIF6 on the Electrical Conductivity of Cryolite-Alumina Melts", J.
Electrochem. Soc., 116, 1969, pp1381-1383
73 Lewis, R. A., "Technical Fundamentals of the Aluminum Reduction Cell

Process", Reduction Division Technical Manual I, Kaiser Aluminum
Internal Publication, 1973
74 Haupin, W.E., "Overvoltages Determined by Nonlinear Regression

Analysis of Cell Volts vs. Current Density", AIME Light Metals, 1980m
pp183-190
73
75 Dewing, E.W., "Liquidus Curves for Aluminum Cell Electrolyte v.
Representation by Regression Equations", J. Electrochem. Soc., 117,
1970, pp780-781
76 Thonstad, J. and Roiseth, S., Proceedings of 3rd ICSOBA Conference,

Nice, 1973, p657
77 Haupin, W.E. and Frank, W.B., "Comprehensive Treatise of

Electrochemistry", eds. Bockris, Conway, Yeager and White, Plenum
Publishing Corp., New York, 1981, p308
74
4.0 Development of a Dynamic Model
A conceptual model of the aluminium reduction process is shown in figure
4.1. The interconnections between the functional blocks represent the flow
of mass, energy and information. Each block represents a set of equations
describing the relationships between the inputs and outputs of that block.
These relationships take the form of differential and algebraic expressions
derived from analysis of published work, plant data and mathematical
modelling techniques.
4.1 Reactor Modelling
Electrochemical cells can be analysed in the same way as any chemical

reactor. The equations describing the process fall largely into two types:
balance equations (materials and energy) and rate equations (reaction
rate, heat and mass transfer etc). A general mass balance would be
written
Rate of accumulation = Rate in - Rate out + Rate of reaction
For a stirred tank reactor with a number of streams entering and leaving
the vessel this can be written
~dCj - fC j dV (1)
V.L..J- -
j=1 dt j=1 dt
where V = volume of the reaction mixture

C·J = concentration of species j in the vessel
Qk = flow rate of stream k into the vessel
Cfk ,J· = concentration of species j in feed stream k
Q1 = flow rate of stream lout of vessel
R·J = rate of reaction for species j
The mass balance for an electrochemical cell differs primarily in that there
are both chemical and electrochemical reactions to be considered. The
reaction term in equation (1) therefore becomes
75
LI.~--------~~~~=::":'--------~
Bath temperature
Freeze melt rate Freeze
r-1 Sludge
Formation
Anode
change
AI product
Breaker
action 1: ) -\ Dissolution I Net
-( 1: 1 -I
Electrolyte
mass
balances
I
r---1 Heat
losses I
.....f - - -
Dissolution
Rate AI 2 0 3
Stochastic
disturbance AI 2 0 3 concentration
D1 concentration estimated
Anode beam
position 1: )
t -
Icond~ctivity
Electrolyte ~
rop 1:
Cell R =V / I
volts Energy = 12 R
Pot heat
balance Bath
temperature
Stochastic
disturbance
D2
Electrolysis Electrolysis
charge heat
balance balance
Line current
Figure 4. 1 Conceptual Model of the Aluminium Reduction Cell

m m
- I j=1
Rj (chemical) + I
j=1
Rj (electrochemical) (2)
In the aluminium smelter, the feed streams into the cell are the feed rates
of alumina, cryolite and aluminium fluoride into the electrolyte. These
flowrates are not continuous and the cell can be treated as a semi batch
reactor where the flowrates, Ok, and the feed concentrations, Cfk,j' vary
discontinuously with time. The main stream out of the reactor, 0 1, is the
periodic tapping of aluminium from the metal pad.
In addition to the heat and mass balances for the reactor, the model will
also include many algebraic equations describing the relationships
between parameters in the model, for example, the dependence of density
on the composition and temperature of the electrolyte. The remainder of
this chapter will describe in detail the modelling of each block in the
conceptual model and the equations used therein. This represents the
physico-chemical model of the aluminium cell. The model of the control
system to be applied in the operation of the cell will be considered in the
following chapter.
4.2 Process Inputs
The process has three primary inputs:-
4.2. 1 The breaker action
The crust breaker movement is controlled by pneumatic pumps, activated

either manually or by the control computer. The pumps are either off or on
and movement is controlled by switching the pumps for finite periods of
time. The breaker can be raised or lowered into the crust, forcing a hole in
the alumina and pushing a portion of alumina rich crust into the cryolite
melt. The breaker action affects both the heat and mass balances in that
immediately following a breaker action there is a sudden increase in the
mass of alumina in the bath followed by an increase in the concentration
of dissolved alumina. A loss of heat also results both from heating the
alumina to the bath temperature and from the exposed bath where the
breaker has broken through the crust.
76
4.2.2 Anode beam position
The anode beam position is controlled in the same way as the crust
breaker by pneumatic pumps. Input is as a number of seconds activating
the raise or lower pumps. The anode beam position governs the anode to
cathode distance (ACO) and hence the voltage drop across the
electrolyte. It also affects the mass balance as it is one of the parameters
that affects the reoxidation of aluminium.
4.2.3 Line current
All cells in series draw the same current. Anode effects in other cells in a
line can cause current fluctuations in individual cells. Current fluctuations
are generally small and in this model the current will be assumed to be
constant throughout the simulation. As well as determining the power
input to the cell by the ohmic drop across the cell, the current determines
the rate of the electrochemical reactions.
4.2.4 Anode change
The replacing of spent anodes is a manual operation that generally occurs

at regular time intervals. Anodes may also be removed out of schedule if
there is a problem such as spiking. The input is the duration of anode
change. The changing of an anode represents a major disturbance to both
the heat and mass balance in the cell.
4.2.5 Stochastic disturbances

Two stochastic disturbances are assumed to enter the system and
represent a random variation in parameters in the model. The disturbance,
01, represents the uncertainty in the amount of alumina fed into the
electrolyte by the breaker action. O2 represents the uncertainty in the
anode beam position due to leakage from the air lines supplying the
pneumatic motors etc. The stochastic disturbances are necessary in the
model to fully test the control strategies to be implemented on the cell.
77
4.3 Alumina Dissolution
It has been shown in the previous chapter that the dissolution of alumina
is dependent upon the method of feeding. This is due to the formation of
agglomerates and sludge in the electrolyte. Laboratory experiments have
differentiated between two extremes of dissolution, that from finely
dispersed particles of alumina and that from alumina agglomerates and
sludge. In an industrial cell dissolution will be by a combination of effects.
This is reflected in the models used by Blake[1] and Entnen2] in which
alumina is assumed to exist as dissolved ions, a suspension of small
particles and sludge. The models differ in that Blake assumes a sequential
mechanism from sludge to suspension to solution whilst Entner assumes
a parallel mechanism from sludge to solution and suspension to solution.
Blake assumes a proportion of alumina to enter each state on addition to
the electrolyte whilst Entner assumes a settling of suspended alumina to
sludge.
The sequential model for dissolution must assume the stripping of alumina
from sludge into suspension by the movement of electrolyte across the
surface of the sludge. This is more applicable to sludge at the side walls
than to large agglomerates floating on the metal pad and is therefore a
more realistic model for side break cells in which large amounts of sludge
are present at the sidewalls than in half break and point feed cells.
A model for the settling of alumina in the cell will produce an alumina
concentration distribution but will increase the number of calculations
required for solution. The accuracy of the model will depend upon the
number of zones used and the assumptions for the settling of alumina.
This in turn will depend upon the geometry and hydrodynamics of
individual cells. If a uniform distribution of alumina is assumed then this
model approximates to that used by Blake.
In this work, alumina will be assumed to exist in three states in the

electrolyte and these will be termed solution, suspension and sludge. In
solution alumina will be assumed to exist as A13+ and 0 2- ions. In practise
dissolved alumina is thought to exist as AI-O-F complexes. It is the
individual ions that are of interest in the mass balance, however, since it is
the oxide ion that is oxidised and reacts with the carbon anode to form
CO2 gas. Measurement of the concentration of alumina in the electrolyte
78
detects the presence of the oxygen atom by precipitation of alumina and
so empirical relationships do not depend upon the form that dissolved
alumina takes.
In line with experimental results[3] alumina in suspension will be assumed

to exist as alumina particles and small agglomerates suspended in the
electrolyte. Sludge will refer to larger agglomerates and alumina deposits
on and around the aluminium, not the bottom crusts and deposits
sometimes found under the metal pad.
The assumption that a proportion of alumina added to the cell goes

directly into solution used in Ek's model[4] is only valid when the rate of
dissolution immediately after addition is very great. This will be the case
when the alumina concentration at addition is well below the saturation
concentration of the electrolyte and the driving force for dissolution is
great. If the concentration is close to saturation this assumption will no
longer be valid. This situation can occur if the cell is overfed, perhaps due
to an inadequate control system. As the primary purpose of this work is to
investigate control strategies, it is desirable to be able to simulate this
condition. For this reason, all alumina added to the cell will be assumed to
enter only as suspended particles or sludge.
There is general agreement that the rate of dissolution can be described

by a first order model at least at higher concentrations of alumina. The first
order model will approximate to the zeroth order model where the
saturation concentration is much greater than the actual concentration.
The first order model is therefore more generally applicable and shall be
used in this work. The rate of dissolution is then given by equation (3)
(3)
where r = rate of dissolution of alumina g cm- 2 S-1
A = contact area for dissolution cm 2

ko = reaction rate constant cm S-1
E = activation energy kcal mole- 1

R = universal gas constant kcal mole- 1 K-1
To = reference temperature OK
79
T = reaction temperature oK
*
c = saturation concentration of alumina in melt g cm-3
c = concentration of alumina in melt g cm-3
Values for the parameters in the above equation have been determined by
a number of sources and have been summarised by Asbj0rnsen[51. The
data from Kachanovskaya will be used in the simulation, the values of the
parameters being ko = 0.00124 cm S-1, To = 1383 oK and
E = 18.153 kcal mol- 1 . For simplicity, alumina dissolving from both sludge
and suspension will be assumed to obey the same rate expressions with
the same values for kinetic parameters. The dissolution rates will be
different due to differing areas of contact between the electrolyte and the
alumina as sludge and in suspension
The individual dissolution rates will therefore be
rdiss(suspn) = Asuspn· K. (c* - c) (4)
rdiss(sludge) = Asludge . K. (c* - c) (5)
where
(6)
suspn refers to alumina in suspension

sludge refers to alumina as sludge
The contact area for alumina in suspension will be less than the specific
area of the alumina as measured in a typical analysis due to the presence
of alumina/bath agglomerates. Sludge will tend to form piles on the bottom
of the cell and the contact area will be further reduced. The contact areas
will be based upon the specific area for the particular alumina used but will
be fitted to produce the desired dissolution curve. This will allow the
simulation of the effects of different types of alumina and different cell
geometries allowing the formation of different sized agglomerates.
Alumina fed to the bath contains an amount of Na20 which reacts with
AIF3 in the electrolyte
80
Although the proportion is small, typically 0.5°10 Na20 by weight, over a
period of time it has a noticeable effect on the weight ratio NaF/AIF3 in the
electrolyte. For simplicity, all alumina in suspension and sludge will be
assumed to contain Na20 in the same proportion as the alumina feed. The
dissolution rate of Na20 into solution will be assumed to be proportional to
that of alumina and given by
rNa20diss = °loNa20. ( rdiss(suspn) + rdiss(sludge) ) (7)
where rNa20diss = net dissolution rate of Na20 into solution

°loNa20 = wt°lo Na20 present in alumina fed to the cell
In many cases the alumina is also blended with material captured in the
l
exhaust scrubbers. This Icatch consists predominately of NaAIF4 from
particulate fluorides and AIF3 from the chemisorbtion of gaseous fluorides
onto A1 20 3. It may therefore be considered to be a mixture of NaF and
AIF3 and the dissolution rates will be determined in the same way as that
for Na20.
rAIF3diss = °loAIF3· ( rdiss(suspn) + rdiss(sludge) ) (8)
rNaFdiss = °loNaF. ( rdiss(suspn) + rdiss(sludge) ) (9)
Again the actual amounts of each compound in the feed are small but
significant with the total fluoride content of a blend being typically 0.5-1.0
wt°lo.
4.4 Electrolyte Mass Balance

The electrolyte mass balance block as shown in figure 4.1 may be divided
into a number of separate processes. These are shown in figure 4.2.
In order to reduce the number of calculations required, and hence the

simulation time, the bath will be considered to be a well mixed system. All
species in the melt will be considered to be uniformly distributed
81
Bath
Volatilisation
Alumina Mass ~
Dissolution Balances
Back
Reaction
Figure 4.2 Electrolyte Mass Balance Processes

throughout the melt. As the electrolyte is constantly in motion due to
electromagnetic stirring of the metal pad this is a reasonable assumption.
4.4. 1 Bath volatilisation
Haupin's model for bath volatilisation[6] includes gaseous losses by

chemical reaction and particulate losses by volatilisation and entrainment.
The mechanisms for the first two losses have been widely investigated
and the relationships appear to be reasonably accurate, with coefficients
in that for gaseous fluoride evolution being fitted to Alcoa test data. The
relationship for entrained electrolyte, however, is derived by difference
from experimental data and the predictions of the other models. It
therefore contains all the errors of measurement and the predictive error
of the models, and, according to Haupin, will predict negative
entrainments under certain conditions. As this would necessitate a
discontinuity in the model and the amount of entrained electrolyte under
normal conditions may be expected to be small, losses due to entrainment
will not be included in this work.
The relationships derived by Haupin express the bath losses in terms of

fluoride evolution as kg F/kg AI produced. For the complete dynamic mass
balance on all species required here the relationships must be rearranged
to express the rate of loss of individual electrolyte components through
time.
4.4.2 Gaseous emissions

The model for gaseous emissions only includes losses due to the
hydrolysis of bath to HF. The reaction responsible for these losses is given
by Haupin as
[2]
If the water for this reaction is assumed to be an addition to the electrolyte

then this will cause an increase of oxide ions in the bath and a decrease
of fluoride ions.
The rate of fluoride evolution is given by Haupin in kg Fluoride per metric

ton aluminium produced
82
F
G
= 2310000
%CE
_ 955000. R .[
P b
( 8444)]
exp 7.553 - T .
b
0/oH 20 in AI 20 3 + %H in anOdeS]~ .
[ 37.44 21.5
(10)
where = gaseous fluoride evolved per metric ton aluminium

produced kg FI tonne AI
%CE = current efficiency as percentage
Rb = weight ratio NaF/AIF3
Pb = barometric pressure kPa
CAI203 = concentration of alumina in bath wt%
C~~:03 = saturation concentration of alumina wt%
The production rate of aluminium is given by
%CE
.- moles AI S-1
100 3·F
%CE 27
--.-.
100 3· F 1x 1 0 6
0/oCE·I·9 tonne AI S-1 (11 )

8
F·1 x 10
where rAI = rate of aluminium production

%CE = current efficiency %
I = total current for electrolysis A
F = Faradays constant C mol- 1
The rate of evolution of gaseous fluoride is then given
83
dMF(Q)
dt
= %CE ._1_. 27 .(2310000 955000 [( 8444)]

100 3·F 1x10 6 %CE - P ·R b • exp 7.553- T .
b
%H 2 0 in AI 2 0 3 + %H in anOdeS]~ .
[ 37.44 21..5
exp( 0.44199 - 3.1733R b + O. 78127R~). C:~~03

C ]-0.462 J
[ (12)
AIP3
The rate of gaseous fluoride evolution is the rate of reaction [2]. By

stoichiometry the rate of addition of oxide ions to the electrolyte by this
reaction is
dMO(g) 1 dMF(g) Wo
- _e . _
gO s-1 (13)
dt 2 dt WF
dMO(g)
where dt = rate of addition of oxide to bath by gaseous
evolution of fluoride
Wo = atomic weight of oxygen = 16 g mol- 1
WF = atomic weight of fluorine = 19 g mol- 1
4.4.3 Particulate emissions by volatilisation
As with gaseous emissions, Haupin derives a relationship for the evolution

of particulate emissions in terms of kg fluoride per tonne aluminium. Using
the information in his model the rate of volatilisation of electrolyte may be
determined.
Vapour over electrolyte has been demonstrated to be composed of

NaAIF4 in equilibrium with its dimer, Na2AI2Fa. Assuming that the pot
gases are saturated with these molecules at their vapour pressures, the
mole fractions of the monomer and dimer in the gas are
(14)
84
(15)
where m,d refer to the monomer and dimer respectively

x = mole fraction of species in pot gases
p = vapour pressure of species
P b = barometric pressure above electrolyte
The maximum rate of evolution of pot gases is the rate of production of

gases by electrolysis at the anode. If the gases evolved in a cell are
assumed to be CO 2 under normal operation and CF 4 during anode effect
then the rate of evolution is given
rpot gas mol S-1 (16)

4·F
and the rates of volatilisation
___ .Pm mol NaAIF4 s-1 (17)

4·F Pb
and
-_.& (18)
4·F Pb
where rm and rd are the rates of volatilisation of monomer and dimer

respectively. The total rate of material volatilised from the electrolyte
expressed as NaAIF4 is then
dM vo'
dt
g S-1 (19)
where WNaAIF4 = molecular wt of NaAIF4 = 126 g mol- 1
85
The vapour pressure of molten cryolite with additives has been extensively
studied and there is good agreement among recent investigators. The
data has been combined by Haupin to produce the following formula
which will be used in this work
kPa (20)
where
A - 7101.6 + 3069.7· Ratio + 635.77· Ratio 2 + 51.22· (%LiF)
- 24.638. Ratio. (%LiF) + 764.5· (%A1 2 0 3 )

[1 + 3.2029· (%AI 2 0 3 )]
B - 7.0184 + 0.6844· Ratio - 0.08464· Ratio 3 + 0.01 085· (%LiF)

1.1385· (0/oAI 2 0 3 )
- 0.005489· Ratio· (%LiF) + [ ( )]
1+3.2029· 0/oA1 2 0 3
+ 0.0068· (0/oCaF2 )
where % = wtO/o of species in electrolyte

Pv = vapour pressure above electrolyte
The total vapour pressure above the electrolyte is also the sum of the
partial pressures of the monomer and dimer.
Pv = Pm + Pd (21 )
The equilibrium constant for dimerization has been correlated from mass
spectrographic measurements[7] and is given by Haupin as
= exp (
-21414
T +15.6
J (22)
where Kd = equilibrium constant

T = temperature oK
combining equations (21) and (22) and rearranging to eliminate Pd gives
86
_ [-1 + (1+4·K d.PV)O.5]
Pm (23)
2·K d
introducing equations (21) and (23) into (19) and rearranging gives
dM VO'
dt (24)
and in terms of oxide ions, sodium fluoride and aluminium ions
dMO G 1 dMFG,net .Wo

- -. - gO S-1 (25)
dt 2 dt WF
dMNaFVO' dMFvo' W NaF

- g NaF S-1
dt dt (26)
WNaAIF4
dMAI~~, dMFvo' W A1
- g AI S-1 (27)
dt dt WNaAIF4
dMO G
where = rate of addition of 0 2 - ions to electrolyte by the
dt
evolution of HF gas g S-1
dMNaFvO' f
-----'-~ = rate 0 volatilisation of NaF g S-1
dt
3+
dMAl vo,
= rate of volatilisation of A13+ ions g S-1
dt
The pot gases from all cells are collected by an exhaust system and
passed into a baghouse where particulate fluorides are filtered and
gaseous fluorides are chemisorbed onto alumina. The product, termed
fines, is added to the alumina fed to the cells.
87
4.4.4 Aluminium dissolution and back reaction
All the models proposed for the rate of the back reaction of aluminium
require the estimation of a number of parameters. In general, these
parameters will differ from cell to cell and the ability to tune the model to
the real cell will depend upon the number of parameters that must be
estimated and the complexity of the model. The complexity of the model
also determines the speed of calculation in dynamic simulation.
The models of Robl[8] and Evans[9] may be capable of estimating the rate
of reaction in different sections of the cell but the difficulty in estimating the
parameters required reduces the accuracy under dynamic conditions. The
complexity of these models also makes them unsuitable for dynamic
simulation.
The assumption that the bulk of the electrolyte is well mixed leads to a
uniform distribution of all species. Under these conditions the mixed rate
models for the back reaction of aluminium proposed by Haupin and Robl
reduce to the dissolution rate controlled model of Liliebuen[10]. This model
has shown reasonable agreement with industrial cells when used for the
calculation of steady state current efficiencies and shall be used in this
work.
The loss of current efficiency in the cell is assumed to be by the reaction
[3]
which is assumed to occur instantaneously between dissolved species.

The rate of reaction is therefore controlled by the dissolution rates of the
reactants into the electrolyte. These rates are given by Lillebuen when the
concentration of CO 2 in the bulk is assumed to be zero
(28)
where r = rate of dissolution of species
88
A = interfacial area of species and electrolyte
k = dissolution rate constant
C = concentration of species
C* = saturation concentration of species
suffix AI refers to aluminium
CO 2 refers to carbon dioxide
These rate equations have been used by Lillebuen to calculate steady

state current efficiencies by equating the rates of dissolution of each
species to the rate of reaction. The current efficiency has then been
calculated from the rate of dissolution of aluminium and it has been
implied that this is the rate controlling step. In the dynamic situation the
assumption that the concentration of dissolved carbon dioxide , Ceo2' is
zero means that all CO 2 that dissolves in the melt must be consumed by
the reaction. The rate of the back reaction is therefore controlled by the
rate of dissolution of CO 2 only. From the stoichiometry of reaction [3] the
rate of reaction of aluminium dissolved in the electrolyte is therefore
mol AI S-1 (29)
It is still necessary to calculate the rate of dissolution of aluminium for the

mass balance on aluminium in the electrolyte.
The models presented for the dissolution of aluminium based upon the
hydrodynamics of the system (Robl, Evans) are complex and require the
estimation of a large number of unknown quantities. The advantage of
these models is offset by the lack of any data relating parameters in the
equation to operating conditions.
Of the models based upon dimensionless parameters only that of

Vasiliadis is for a similar physical system to the aluminium cell. Although
this relationship does not include a complex hydrodynamic model it does
include the effect of gas bubbles at the anode and all the parameters
required are obtainable from literature or by direct measurement.
Furthermore the dissolution rate is directly related to parameters known to
effect the current efficiency of a cell. For these reasons Vasiliadis' model
for the rate of dissolution of aluminium will be used in this work. The
dissolution rate constant for aluminium in the electrolyte is
89
- 2.147x10- ( T
11 i. T )1.42
.(1+313.3V 2.5 ).
(
~::
J
O 33
.
·D~~7 (30)
where kAI = dissolution rate constant m S-1

= anodic current density A m-2
T = electrolyte temperature K
f = anode to cathode distance m
V = max velocities in the metal pad m S-1
J.l = viscosity of electrolyte Ns m-2
P = density of electrolyte kg m-3
DAI = diffusivity of aluminium into electrolyte m2 S-1
The dissolution rate constant for carbon dioxide has been proposed by
Haupin as
Dco 1. 703 - 0.7630 ·lnReco

_-=-2 • 2 • Re o.s . Sc o.s (31 )
Dco 2 1- 0.6272 ·In Re co2 CO 2
where = diffusivity of carbon dioxide in electrolyte

= carbon dioxide bubble diameter
= Reynolds number for carbon dioxide boundary layer
_ Ve' Pe' Dco 2
/le
Sc = Schmidt number for carbon dioxide in electrolyte
__-,-P....;:..e_
Pe . DC0 2
Ve = velocity of the electrolyte

Pe = density of electrolyte
/le = viscosity of electrolyte
Haupin does not, however, state how this relationship was determined
and so no comment may be made upon its validity. There is as yet no
data relating bubble diameters to operating conditions or the geometry of
the cell. As the current efficiency is determined from the dissolution rate of
carbon dioxide these relationships must be known in order to simulate the
effect of the parameters. Until such data is available Haupin's relationship
has limited use in dynamic simulation.
90
Lillebuen proposed that the dissolution rate constants for aluminium and
carbon dioxide were equal based upon the assumption that the interfacial
area for carbon dioxide dissolution is ten times that for aluminium and the
ratio of kc02 'Ac02 to kAI'AAI is 0.1. Vasiliadis, however, found an average
value of this ratio to be 0.06 and hence kAI / kco = 0.6.
2
The interfacial area for carbon dioxide dissolution has since been related
to the concentration of alumina in the melt but unfortunately to no other
parameters such as temperature. Using Haupin's relationship however
(0.02723+ 1.003Wt-0.3208We -0.05874Wt 3

+ 0.03268Wt 4 + 0.00181Wt 5 )
--------(----------------~--~------,I cm 2 (32)
1- 0.4715Wt + 0.06247We) 0
where Ac02 = interfacial area of CO 2 gas / electrolyte

Wt = wt% of alumina in solution
10 = current associated with the oxide reaction
Valid between 1 and 8 wt% alumina
it can be seen that the ratio of areas will vary with alumina concentration
and anode current as the aluminium pad area remains approximately
constant. The ratio of mass transfer coefficients will therefore depend
upon the average operating conditions. In this work, due to the lack of
carbon dioxide bubble data, the dissolution rate constant for carbon
dioxide will be determined
(33)
where f is a fittable constant that may be adjusted to account for the

average operating conditions in the cell.
4.4.5 Current efficiency during anode effect

All the mathematical models in literature have dealt specifically with loss of
current efficiency under normal operating conditions where the reaction is
assumed to be between aluminium and carbon dioxide. During anode
effect, however, carbon dioxide is no longer produced at the anode and
since there is assumed to be no accumulation of carbon dioxide in the cell
91
this reaction is no longer possible. If this is the only mechanism for current
efficiency losses then during anode effect an efficiency of 100% would be
obtained. General operating experience on central point feeding cells has
indicated high current efficiency at low alumina concentrations ie close to
anode effect although this could be due solely to the reduction in bubble
surface area at low alumina concentrations as shown by Haupin1s
relationship and not the presence of an alternative reaction. The current
efficiency during anode effect in industrial cells cannot be determined by
gas analysis and the Pearson and Waddington equation because the gas
contains species other than CO and C0 2 J11]
Assumption that loss of current efficiency is only due to a reaction with

CO 2 introduces a discontinuity at anode effect. The presence of a
fluorocarbon bubble adhering to the anode will prevent aluminium in the
melt from physically reaching the surface and so, if the current efficiency is
not to be 100% , an alternative reaction must account for this loss. In this
work, the anode gas produced during anode effect is assumed to be
purely CF4, The back reaction for aluminium during anode effect will be
assumed to occur between dissolved aluminium and CF 4 and it will be
treated in the same way as that during normal operating conditions with
the reaction being
AI(soln) + 3/4CF4(soln) ~ AIF3(soln) + 3/4C(soln) [4]
and occurring between dissolved species. Carbon particles are found in

the electrolytes of industrial cells where they burn on the surface of the
bath although these could come from break down of the anode itself.
The rate of dissolution of CF4 into the melt is then given
(34)
and by stoichiometry the rate of consumption of dissolved aluminium by

reaction [4] is
(35)
92
Due to the lack of data available for the dissolution of CF 4 into molten
electrolyte the dissolution rate constant and the saturation concentration
of CF4 will be taken to be the same as the values calculated for CO
2
dissolution. As fluorocarbons cause dewetting of the anode surface and
coalesce to form a single large bubble, the interfacial area between
electrolyte and CF 4 gas will be approximately equal to the surface area of
the anode covered by the gas. Defining a fractional coverage of the anode
surface by CF 4 as <I> gives the interfacial area for CF4 dissolution as
(36)
where AAnode is the effective area of the anodes for electrolysis.
In the build up to anode effect both oxygen and fluoride ions are oxidised
and during this period aluminium will be consumed by both the above
reactions. The net rate of consumption of dissolved aluminium is then
4.4.6 Current efficiency losses at the cathode
Current efficiency is also lost by the co-deposition of undesired metal ions

eg sodium at the cathode. This decreases the partial current density
associated with the reduction of aluminium ions and hence the rate of
reaction below the theoretical maximum. Some measure of this effect can
be determined from the purity of the aluminium produced. Crudely, the
purity of the aluminium will be inversely proportional to the amount of co-
deposition. Under normal conditions in a particular cell the concentration
of sodium in the aluminium pad will remain approximately constant and it
will be assumed that the current efficiency of the reduction reaction of
aluminium is constant. This effect will, in any case, be far outweighed by
the current efficiency losses already discussed above.
By definition the current efficiency of an electrochemical reaction
In- + ne ~ J (38)
at an electrode is
93
(39)
where CE j = current efficiency for reaction of species j

Ij = current associated with the electrolysis of j
It = total current through the electrode = L ij
The rate of reduction of aluminium at the cathode will therefore be
- CEAlo-- (40)
ZAl oF
where = total current flowing through the cell

CE AI = cathodic current efficiency for the reduction of AI
4.4.7 Overall current efficiency
The overall current efficiency is defined as the ratio of the actual rate of
production of aluminium to the theoretical rate and differs from that
determined by the Pearson Waddington equation in that it includes effects
other than chemical reaction of aluminium at the anodeo Here, the current
efficiency losses are assumed to be due to back reaction and cathodic
losses. The net rate of aluminium production is therefore
rate of production at cathode - rate of back reaction
(41 )
The overall current efficiency of the cell is then given
CE =
CE _ rback (42)
- AI r.
t
where the theoretical rate of production, rt , is determined from Faraday's

law and the total current through the cell
I (43)
-
ZAl oF 3F
94
4.4.8 Mass balances
In order to fully describe the system a mass balance must be performed

for each species present in the electrolyte and overall for the molten bath
and metal pad. As the bath is assumed to be well mixed the differential
equations need only to be derived with respect to time and not space. In
this block it is also necessary to calculate the concentrations of species in
the electrolyte. In this work the electrolyte is assumed to consist of
alumina, sodium fluoride, aluminium fluoride, calcium fluoride and
aluminium. For simplicity all species are assumed to dissociate into simple
Ions.
The general form for a mass balance on species j is then
Accumulation Mass of j Mass of j Mass of j

of added removed consumed
species J to system from system by reaction
where the reactions may be electrochemical or chemical. For a system

with a number of streams entering and leaving this may be expressed
- R.J (44)
where Mj = mass of species j in the system

C·J = concentration of species j in the system
Qk = flow rate of stream k into the system
= concentration of species j in feed stream k
QI = flow rate of stream lout of system
Rj = net rate of reactions for species j
The mass balances for individual species in the aluminium reduction cell
are developed in the following sections.
4.4.9 Material in suspension and as sludge

The alumina fed to a cell is not pure but contains significant quantities of
Na20, NaF and AIF3 the latter two compounds being added to the feed by
blending the alumina with material captured in the baghouse. The actual
quantities are small compared with the mass of alumina, typically less
95
than 2.0 wt% total, but have a marked effect upon the operation of the cell
and therefore must be included. For simplicity the rates of dissolution of
each compound are considered to be proportional to the rate of
dissolution of alumina. It is therefore assumed that all sludge and
suspension has the same chemical composition as the feed and the mass
balances will be performed as on a homogenous material.
The mass balances on material in suspension and as sludge are
In: material added as a feed by the crust breakers

material added by action of changing an anode
Out: material dissolved into the electrolyte
In this work alumina added to the electrolyte is assumed to become

sludge or suspension by a constant proportionality factor. The mass
balances may be expressed mathematically
dMsuspn (45)
dt
dMs,udge
- s f' [dM
dt
feed
- dM
dt
ac
- rdiss(sludge) ]
(46)
dt
where = fraction of feed that becomes sludge
dM feed = rate of addition by crust breakers

dt
dM ac = rate of addition due to anode change
dt
Alumina additions by the crust breakers and anode change are not
continuous and it will be assumed that any alumina introduced to the
electrolyte will instantly enter the sludge and suspension states. The
addition rates may therefore be represented by a series of impulses with
varying period and height. The period represents the frequency of
operation of the crust breakers or anode change and will be set by the
control system. The height represents the amount added to the cell and is
subject to a random proportionality factor to simulate the uncertainty
96
between the actual mass added and the intended mass. The mean and
variance of the amount of alumina added to the cell may be measured by
sampling a working cell. The random variable may then be generated from
a distribution function based upon these variables. An example of the form
the rates of addition might take is shown in figure 4.3.
4.4.10 A/3+ ions in solution
In: By dissolution of AI 20 3 from sludge and suspension

= 2 x molar rate of dissolution AI 2 0 3 x atomic weight AI
By dissolution of AIF3 from sludge and suspension
= molar rate of dissolution AIF3 x atomic weight AI
By addition of bagged AIF3 to maintain ratio
= molar rate of addition AIF3 x atomic weight AI
Out: By volatilisation of bath as NaAIF4

Rate of freezing of A13+ ions into frozen sidewalls
Rate of freezing of A!3+ ions onto cold anode
Electrolysis: - Rate of reduction at cathode
Reaction: + Rate of back reaction
expressed mathematically this becomes
dMAI 3+ bag
-_ [ 2 . ( rdiss(sludge) + rdisS(Suspn) ) .
+ r AIF3 dlss
+ dM
dt ]. WAI
dt
dMAI~~1 dMAI~+ dMAI!~ _ CE. .W + ~ (47)
- d AI F AI back
dt dt t ZAI'
where dM bag = molar rate of addition of bagged AIF3

dt
dMAI;+ = rate of transfer of A13+ ions into frozen

dt
sidewalls, +ve for freezing
dMAI~~ = rate of transfer of A!3+ ions into frozen

dt
layer on cold anode, +ve for freezing
97
-25 -20 -15 -10 -5 o 5 10 15 20 25
Percentage Overfeed
Figure 4.3 Random Variation on Mass of Alumina Fed to a Cell

W A1 = atomic weight of aluminium = 27 g mol- 1
ZAI = charge transferred in reduction of AI = 3
4.4. 11 ()2- ions in solution
In: By dissolution of AI 20 3 from sludge and suspension

= 3 x molar rate of dissolution AI 20 3 x atomic weight 0
By dissolution of Na20 from sludge and suspension
= molar rate of dissolution Na20 x atomic weight 0
Out: Rate of freezing of 0 2- ions into frozen sidewalls

Rate of freezing of 0 2- ions onto cold anode
Electrolysis: - Rate of oxidation at anode
Reaction: + Formation of AI 20 3 by back reaction with CO2

= 0.5 x molar rate of dissolution of CO 2 x atomic weight 0
+ Formation of AI 20 3 by HF reaction
This is expressed mathematically as
dM0 2 -
= [3. (rdiSS(SIUdge) + rdiSS(SUSPn)) + rNa20diss ]. Wo
dt
dMO!~ -CE I.W

dt dt 0 zo. F 0
(48)
dM0 2f -
where ------.:.- = rate of transfer of 0 2- ions into frozen
dt
dM0 2 -
_------=a:.::--c = rate of transfer of 0 2- ions into frozen
dt
layer on cold anode, +ve for freezing
Wo = atomic weight 0 = 16 g mol- 1
98
4.4. 12 Aluminium metal in electrolyte
Aluminium dissolves from the metal into the electrolyte where it can react
with dissolved CO 2 or CF 4 , The dissolved aluminium is assumed to be non
ionic. The mass balance is then
In: Rate of dissolution of aluminium from metal pad
Reaction: Rate of back reaction of aluminium with CO 2 and CF 4
Mathematically this is
dMAI
- rAldiss - rback (49)
dt
4.4. 13 Aluminium in the metal pad
The metal produced in an aluminium cell contains a small percentage of

impurities such as silicon and sodium. In this work the only impurity to be
considered is sodium which is assumed to exist in the aluminium pad at a
constant concentration. The actual amount is however, very small with the
purity of aluminium produced being typically greater than 99.9%. The
amount of sodium lost from the melt is therefore considered insignificant
and will not be included in the mass balances. A mass balance on the
aluminium in the metal pad will therefore constitute a mass balance on the
whole of the metal pad.
The mass balance is
Out: Aluminium periodically tapped from the cell

Rate of dissolution of aluminium into the electrolyte
Electrolysis: + Rate of reduction of A13+ at the cathode
Mathematically this is
dMAlpad dMAl tap I (50)

----rAldiss +CE A1 · ,WAI
dt dt zA,· F
99
dMAl tap
where dt = rate of metal tapped from the cell
4.4. 14 Sodium fluoride
The sodium content of the cell is assumed to be independent of

electrolysis although strictly a term should be included to account for loss
of current efficiency at the cathode. The actual mass of sodium removed
from the electrolyte by this means is very small however and will be
considered negligible compared to other terms in the mass balance.
In this work all sodium in the electrolyte is assumed to be due to the

presence of NaF. The concentration of individual sodium ions is not
required because all the relationships for physical properties require the
sodium content to be expressed as the weight ratio of NaF to AIF 3. The
only other source of sodium ions is the addition of Na20 as part of the
alumina feed. This reacts with AIF3 to form NaF
[5]
At the operating temperature and composition of the cell this reaction will
be fast and the rate determining step will be assumed to be the dissolution
of Na20. Two moles of NaF will therefore be formed for every mole of
Na20 dissolved.
A mass balance on the NaF content of the electrolyte is
In: By dissolution of NaF from sludge and suspension
Out: Rate of freezing of NaF into frozen sidewalls

Rate of freezing of NaF onto cold anode
By volatilisation of bath as NaAIF4
Reaction: + rate of reaction of Na20

= molar rate of dissolution Na20 x 2 x molecular weight NaF
100
dMNaF dMNaFf dMNaF dMNaF
dt
= ~ ,
NaFdlss
-
dt -
dt ac
- dt vol
+ ~Na 2Od'ISS ' 2 , WNaF (51 )
dMNaFf
where dt = rate of transfer of NaF into frozen
dMNaFac
dt = rate of transfer of NaF into frozen layer
on cold anode, +ve for freezing
dMNaFVO'
------=.=- = rate of volatilisation of NaF
dt
W NaF = molecular weight NaF = 42 g mol- 1
4.4. 15 Calcium fluoride
As with sodium ions, the calcium ions in the electrolyte are assumed to be
associated only with calcium fluoride. In practise the alumina fed to the
cell contains a small percentage of CaD which reacts with AIF3 in the melt
to form CaF2. The amount of CaD in the feed is typically a tenth of the
amount of Na20 and as such may be considered negligible. Similarly, the
mass of calcium co-deposited at the cathode is a fraction of the mass of
sodium and will also be assumed negligible. In this model the calcium
fluoride content of the molten electrolyte will therefore only be affected by
the rates of freezing at the sidewalls and on cold anodes.
Out: Rate of freezing of CaF2 into sidewalls

Rate of freezing of CaF2 onto cold anode
dMCaF2 dMCaF2f, dMCaF2ac

, (52)
dt dt dt
dMCaF
where - - - - ' -2f' = rate of transfer of CaF2 into frozen
dt
101
dMCaF2ac
dt ' = rate of transfer of CaF2 into frozen layer
on cold anode
4.4. 16 Alumina in solution
Although alumina dissolved in the electrolyte has been assumed to

dissociate into A13+ and 02- ions the equivalent mass of alumina is
required since the industry uses the weight percent of alumina to describe
the composition of a melt and most empirical relationships require the
alumina concentration on a weight basis.
Since this work assumes that the only oxide containing species dissolved
in the melt is alumina, (other compounds, eg Na20, react to form alumina)
it follows that the mass of alumina is proportional to the mass of 0 2- ions.
From the chemical formula of alumina, A1 20 3, it can be seen that each
mole of oxide is equivalent to one third moles of alumina. The mass
balance on alumina in solution is therefore
1 dM0 2 - WAI203
- -. . (53)
3 dt Wo
where dMAI 2 0 3 = rate of change of mass of alumina in solution

dt
W AI
2
0
3
= molecular weight of alumina = 102 g mol- 1
4.4. 17 Aluminium fluoride

The mass of aluminium fluoride in the electrolyte must be determined as
many of the empirical formulae require either the weight ratio of NaF to
AIF3 or the percentage excess of AIF3 above the mass found in cryolite.
As with alumina, the AIF3 mass balance may be determined algebraically
from the mass balances on individual ions.
This work assumes that cryolite is a mixture of NaF and AIF3 in the mole
ratio 3: 1. Therefore, the only species in the electrolyte that contain the A!3+
ion are AIF3 and AI 20 3 with mole ratios 1 mole A!3+ per mole AIF3 and 2
102
moles A13+ per mole AI 2 0 3 respectively. The A13+ ions associated with
alumina may be determined from the oxide content of the melt and any
remaining A13+ ions must therefore be associated with AIF . The
3
aluminium fluoride mass balance is
dMAIF3
dt (54)
where WAIF3 = molecular weight of AIF3 = 84 g mol- 1
4.4. 18 Overall mass balance on electrolyte
The rate of change of the total mass of the bath is the summation of the
rates of change of mass of the individual species that make up the
electrolyte. That is
dMBath dMNaF + dMAIF3 + dMCaF2 + dMAI 2 0 3 + dMAI (55)

dt dt dt dt dt dt
This balance only considers species dissolved in the electrolyte, ie the

liquid. The mass of bath is required to calculate concentrations, bath
volume and for the heat balance.
4.5 Electrolyte Conductivity
Whilst the models proposed by Fellner appear to predict the conductivity

of the electrolyte with reasonable accuracy, they rely upon data, such as
the molar volume of each species, that is difficult to obtain for the varying
conditions within the cell. The complexity of the models and thus the
increased computation time required make the models unattractive. The
empirical relationship developed by Choudhary is accurate for the
conditions expected in an operational cell and is easy to calculate. The
relationship has been modified by Roger Johnston at the Kaiser Aluminum
Californian Research Center and is used at Anglesey Aluminium on a
daily basis. The following relationship will be used in this work.
103
1 [2.0156 - 0.0207· %A1 2 0 3 - 0.005· %caF2 ]
p
= 1.06· exp + 0.4349. Rb _ 2068.4 (56)
T
The resistance of the electrolyte can be calculated from the conductivity,

the interelectrode gap and the area for current flow.
f
R = (57)
A·p
The area for current flow through the electrolyte will be affected by the
design of the electrochemical cell. A good approximation for the design of
cell used at Anglesey Aluminium has been found to be the assumption of
a two inch shadow around the block of anodes. The effective anode area
is then
(58)
where n = number of anodes

xa = anode width
Ya = anode length
gap= gap between banks of anodes
This relationship assumes that all the anodes are drawing the same
current. This is not the case however when an anode has been removed
from the bath or a cold anode has been introduced to the bath and is
insulated by a layer of frozen electrolyte. The heat balance across the
anodes and the algorithms developed in a later section for the modelling
of an anode change, calculate a fraction of the surface area of each
anode that is effectively available for current flow. In this work, the
effective anode area will be further modified by subtracting the area of
each anode that is effectively insulated. The calculation of the effective
anode area then becomes
104
where fj = fraction of anode able to conduct current
This modification maintains the two inch shadow around each anode.
4.6 Electrolysis Charge Balance: Reaction Modelling
Although electrode reactions vary in their complexity they can be

represented simplistically by the diagram in figure 4.4 in which reactant A
is transported from the bulk electrolyte to the electrode surface where it
undergoes conversion via, perhaps, a number of reactive intermediates to
product B. B is then transported back to the bulk.
The rate of reaction of A at the electrode is given by the Butler Volmer

equation for the reversible reaction
I+~
A + ne ( ) B (60)
f-I
by the equation
zF zF
-a-ll (1-a)-1l
- 1 ' e RT _ e RT
0 (61 )
where = current density associated with the reaction

= exchange current density where rfor = rrev = i+ = L
io is proportional to zF[A]s
= overpotential at the electrode
= current density associated with forward reaction
= current density associated with reverse reaction
a = transfer coefficient
= fraction of free energy of activation affecting L
105
~~~~~~~~~~
As .t------ Bs
reaction via
intermediates
mass mass
transport transport
Ab Bb
Figure 4.4 Schematic Representation of Reactions at an Electrode
The Butler Volmer equation can be rewritten for the individual rates
zF
-a-1J
- ioe RT
(62)
and
(63)
where E = polarisation voltage = C/>m - C/>S - C/>0

[ ]s = surface concentration of reacting species
Usually the conditions at the electrode are well removed from equilibrium
and i+ or _ » Lor +. The overall rate then becomes
(64)
This equation is known as the Tafel equation and is shown schematically

in figure 4.5. It can be shown that many more complex reactions also
exhibit Tafel behaviour and it is a reasonable assumption that the primary
reactions in the aluminium electrolysis do so.
Note
The Tafel relationship for an electrochemical reaction is analogous
to the Arrhenius rate expression for a first order chemical reaction
where the rate of reaction is given
E
r - A·e RT
where A Arrhenius constant

E - Activation energy of reaction
106
'-
f\
f\
I
W
C!:l. '-c::+
I ·2
.....
Q)
~
~ .~
Q)
W
II
........
..
-.-ctS -
Q
~
~
...... 't--
0
c: E
Q)
...... ~
0)
0 .~
a. -0
.~
.....
ct!
E
Q)
..c::
(.)
C/)
LO
~
@
:J
.~
LL
I
•
'AI!SUaa IUaJJ n8
The rate of reaction of A is given by Faraday's law as
z·F (65)
where rA = rate of electrolysis of species A

I = current for electrolysis
Ae = electrode surface area
IA = partial current density associated with the reaction
The mass transport of species to and from the electrode surface IS
brought about by diffusion processes and, if the species is charged, by

ionic migration. In the presence of a supporting electrolyte the latter is
insignificant and will not therefore be considered here. Transport by
molecular diffusion processes is given by Fick1s first law of diffusion for
species A
(66)
dN = rate of transport of A to the electrode surface

where _.:....:...A
dt
kL = mass transfer coefficient
Ae = surface area of the electrode
[A] = concentration of reactant A
b,s are subscripts referring to the bulk and electrode surface
respectively
At steady state the rate of transport of A to the surface will be equal to its
rate of reaction at the electrode
dNA
-- (67)
dt
and, combining with Faraday's Law
(68)
107
From (65) and (66) the following simultaneous equations can be obtained ,
(69)
iA = z F k [Als e-j3E
(70)
which taken together comprise the model for the reaction A ~ B.
Rearranging (69) and (70) and eliminating [Als gives
- [AJ IA iA
----'-'-e -fiE
b z.F.k L (71 )
z·F·k
and
1 1 (72)
+
z·F·k·e -fiE
z·F·k L
This expression models the reaction independent of a knowledge of

surface concentrations at the electrode.
Although the model has been derived for the simple reaction
A+ze ~ B [6]
it applies to any reaction that exhibits Tafel type behaviour, and can be
shown to be valid for many more complex reactions[ 121.
Much work has been done on measuring the polarisation voltage in

aluminium reduction cells, although most of it has concentrated on
determining the variation of voltage with alumina concentration prior to
anode effect, with little consideration of the voltage rise during the anode
effect itself. The primary electrochemical reaction at the anode during
normal operation is the reduction of the oxide ions associated with the
alumina dissolved in the electrolyte.
0 2- ~ 0 + 2e [7]
108
Most workers have fitted the experimental results to Tafel type behaviour
although some have included a concentration overvoltage effect to
accommodate the rise in voltage at low concentrations of alumina. Some
of the relationships derived will be considered here and the parameters to
be used in this work will be determined.
The annual International Course on Process Metallurgy of Aluminium at

Trondheim considers the polarisation voltage in the cell in detail and the
approach taken covers much of the research into the relationship between
polarisation voltage and alumina concentration. It is this work that will form
the basis for comparison with the model presented here.
In the Trondheim model, the overpotential at the anode is considered to

be represented by two parts, the overpotential at the surface, l1SA'
represented by a Tafel type relationship and a concentration overvoltage,
11cA-
1J = l1sA + 1JCA (73)
where
R·T I
11 SA ---·In-
a·z·F io
Io - Y + 8·c b,A1 0 2 3
a - fn (porosity & reactivity of carbon anode)
_ R.T.lni
z·F ie- i
_ [5.5+0.018(T -1323)].A-o.1,[c~:!'203 -0.4]
_ £. (C~·5 - Jl) for a given temperature
cb,A1
20 3
=concentration of alumina in the electrolyte
A = anode area
For a given temperature the overpotential is then given by
I Ie (74)
11 = A·ln-. + B·ln--
ie -i
Io
109
Rewriting equation (74) in terms of the overpotential , 1] -
-
E - Etio'1', gives
.
I =
1 1 (75)
+
z·F·k L
rearranging gives
17 - 1 1 [ z·F·k·i
-·In-· L
]
b a z . F . kL . c b - i (76)
denoting the limiting current, z·F-kL·Cb by ic and z·F-k L as i~ gives
1 [i', i ]
17 - b1 ·In a' icc - i
1 .,
- -.In~ + 2. lni 1 1
+ -·In--
b a b b ic- i
- a + fi·lni + fi.ln-.1_.
Ie -I
- fi·ln + fi·ln~ where a=-f.?·lna' (77)

a'·i c I c -I I-'
Comparison of equations (74) and (77) shows that although the two
models are similar, they differ in several important respects and it is not
possible to match the coefficients in the models and so determine the
parameters required for this work.
The differences between the Trondheim model and the theoretical model
are probably due to the derivation of the Trondheim model. The
Trondheim model is highly empirical and has been derived over a number
of years by amalgamating several empirical relationships as research has
made them available. Whilst this approach can lead to good results over
the specific range of parameters investigated, it may no longer be possible
to relate the derived relationship directly to the underlying theory, as in this
case, bringing the applicability of the relationship under a wider range of
conditions into question. For this reason the theoretical relationship will be
used in this work with the required Tafel coefficients and mass transfer
coefficients being determined from the available empirical relationships for
polarisation voltage.
110
4.6. 1 Determination of Tafel constants and mass transfer coefficient
Tafel coefficients for a wide variety of carbon anodes with varying porosity
have been determined by Zuca et aH13] and figure 4.6 shows the variation
of overvoltage with current density for different carbon types at 965 ce.
The Tafel plots determined by Zuca are also shown on the graph. The
Tafel coefficients for the equation 11 = a + b log i and the range of current
density over which Tafel behaviour was assumed are shown in table 4.1.
The cryolite composition used was similar to that used in industrial cells
with 5°1o alumina, 4°1o calcium fluoride and a cryolite ratio of 1.35.
The experimental data in figure 4.6 clearly shows the effect of mass
transfer at higher current densities with the rise in potential as the reaction
becomes mass transfer limited. It is evident from the graph that the range
over which the Tafel coefficients are calculated includes some effect due
to mass transfer and so the coefficients cannot be used directly in
equation 72.
Assuming that anode effect is initiated when the oxide reaction becomes
mass transfer limited then at this point the partial current density for the
reaction will be given by
I = z·F·k L ·c b (78)
where Cb = the molar concentration of oxide ions in the electrolyte
Rearranging (78) gives
(79)
It is known that in a typical cell at Anglesey Aluminium anode effect starts

when the alumina concentration in the melt approaches 1.5 wt°/o.
Assuming that a typical cell has the following operational parameters
Calcium fluoride 5 wt°lo

Weight ratio 1.25
Temperature 965 cC
Current density 0.7 A cm- 2
111
Anode Porosity Tafel intercept Tafel slope Tafel range
0/0 V V/decade A cm- 2
spectral graphite 0 0.805 0.274 0.02-0.8

baked carbon 8.9 0.784 0.286 0.02-0.8
12.6 0.710 0.250 0.02-0.7
14.73 0.611 0.202 0.03-0.7
Table 4.1 Experimental Tafel Coefficients for Anodes of Varying Porosity

1
o
0.9 o
/':,
0.8 6X
0.7
>
co 0.6
...c
.-
~ 0.5
~o
I o vitreous carbon
a.
a..
Q) • baked carbon
> 0.4
0 o porosity 8.90/0
0.3 x porosity 12.82%
0.2 /':, porosity 14.9%
0.1
0
0.01 0.1 1 10
Current Density Alcrrt2
Figure 4.6 Overvoltages from Zuca

and that the oxide reaction is the only reaction occurring at the anode at
this point, equation (79) gives the mass transfer coefficient for the
reaction, kL' as 3.9x10-3 cm S-1. For simplicity, it will be assumed that kL is
independent of temperature and composition.
Having obtained a value for the mass transfer coefficient, the required
Tafel coefficients, a and /3, can be obtained by solving equation (77) at
two different current densities, with the overpotential being calculated from
the equations derived by Zuca.
The porosity of the anodes used at Anglesey is approximately 14.75 % •
For a baked carbon anode with porosity of 14.73% , Zuca derives the Tafel
equation
11 = 0.611 + 0.202 log i (80)
over the range i=0.05 to i=0.5 A cm- 2 . The anodic overpotential at these
limits is then
110.5 = 0.550 V and 110.05 - 0.348 V
From equation (77)
11 - a + /3 i .J
.In(-. (81 )
Ie -I
Substituting the values for the composition at which the experiment was
carried out and the calculated value of kL into the equation gives ic =
2.3163 A cm-2 . Substituting this value and the values obtained for 11 at 0.5
and 0.05 A cm-2 into equation (80) gives the two simultaneous equations
a - 1.290 /3 = 0.550 and

a - 3.814/3 = 0.348
112
Solving these gives a = 0.653 and f3 = 0.080 and
11 - 0.653 + 0.080.ln( I
752.7,c b -i
J (82)
Equations (77) and (82) are plotted in figure 4.7 for an electrolyte
composition of 5 wtOlo alumina, 4 wt°/o calcium fluoride, cryolite weight ratio
of 1.35 and at 965°C. The Tafel plot derived by Zuca for an anode
porosity of 14.73°/0, equation (80), is also shown.
As would be expected, the equations (80) and (82) give very close results
for the range of current density over which Tafel behaviour was assumed.
As the current density increases, the rate of reaction becomes mass
transfer controlled and the potential rises more rapidly.
There is a very great similarity between the shapes of equations (74) and
(82) although the equation combining 1JSA and 1JCA' (74), produces an
overpotential some 0.15 V less than that measured by Zuca. This could be
due to the reference used in calculating the overpotential.
In addition to 11SA and 1JCA' the Trondheim model also includes an

empirical relationship for a cathodic overpotential derived by Alcoa.
R.T.(1.375-0.125.CR)ln( ic J (83)
1.5·F 0.257
where 1Jet = total overpotential at the cathode

CR = cryolite ratio, mole ratio NaF/AIF3
Ie = cathodic current density
The combined results of equations (74) and (83) are shown in figure 4.7
with the results from equation (82). The expression for the cathodic
overpotential is only valid at cathodic current densities above 0.1 A cm- 2 .
Inclusion of the cathodic overpotential causes the two relationships to
converge at higher current densities but results in a greater divergence in
the Tafel region of the graphs. Information on the reference potential for
the derivation of equation (83) was not available and it is possible that this
could conflict with the reference used by Zuca or those used in the
113
Wt% Alumina 5%
Temperature 975°C
1.2
Ratio 1.35
Calcium Fluoride 4% 1
->---
Q)
o Tafel only
X Tafel + mass transfer
t:, Haupin, sa + ca 0.8
-...o
C)
co o Haupin, sa + ca + cc
>
c
...
o
CO
0.4
-en
~
~X
-o
CO 0.2
a..
0.1 10
-0.2
Current Density ( A I cm2)
Figure 4.7 Overpotential vs Current Density

derivation of 1JSA and 1JCA' thus resulting in the discrepancy between the
models.
Figure 4.8 shows the behaviour of the models to variation in alumina

concentration at a constant anodic current density of 0.7 A cm- 2 . As the
alumina concentration approaches the point at which anode effect would
be expected, approximately 1.5 wt% at this current density, there is a
marked increase in the cell voltage. Only the model derived here exhibits
this behaviour with the effect not occurring in the other models until much
lower concentrations have been reached.
In this work, equation (82) will be used as the model for the anodic
overpotential of the oxide reaction. The effect of current density upon the
polarisation voltage at different alumina concentrations is plotted in figure
4.9. This graph clearly shows that the onset of anode effect occurs at
higher alumina concentrations as the current density is increased as is
experienced in practice.
4.6.2 Reactor modelling

An electrochemical cell is a heterogeneous reactor with the electrodes
acting as two different catalysts, each forming its own set of products. The
mass balance differs from that of a stirred tank chemical reactor in that the
rates of reaction are for both chemical and electrochemical reactions. The
reaction term in equation (1) therefore becomes
m m
- LR j=1
j (chemical) + LR
j=1
j (electrochemical)
(84)
For a normal chemical reaction the kinetic rate expression would take the
form
_ k. [ ]a. [ ]b (85)
R j (chemical)
where k = rate constant for reaction

[ ] = concentrations of reactants
a,b = constants
114
1.2
1.1
Current Density 0.7 A/cm2
1 Temperature 965°C
-
~
Q)
0.9 o
Ratio
Calcium Fluoride
Haupin, sa+ca
1.35
4%
C) 6. Haupin, sa+ca+cc
~ 0.8 o Tafel + mass transfer
0
>
c 0.7
-
0
ca
~
~
ca
0.6
~ 0.5
0.4
0.3
0.2
1 2 3 4 5 6 7 8 9 10
Alumina Concentration (wt°jo)
Figure 4.8 Overpotential vs Alumina Concentration

1.2
1.1
Temperature 965°C
1 Ratio 1.35
-
~ 0.9
(1) o
Calcium Fluoride
o id = 0.5 A/cm2
id = 0.7 A/cm2
4%
C)
X id = 0.9 A/cm2
~ 0.8
0 I:::. id = 1.1 A/cm2
>
c 0.7
0
+:
co
en 0.6
";:
co
~~------X.-- ___- 5<-
~ 0.5
0.4
0.3
0.2
0 1 2 3 4 5 6 7 8 9 10
Alumina Concentration (wt%)
Figure 4.9 Overpotential by Mixed Model vs Alumina Concentration

but for an electrochemical reaction the corresponding equation is
I·J
R j (electrochemical) - (86)
z·F
where Ij = partial current density of the reaction

z = number of electrons transferred in the reaction
F = Faraday's constant
The total current is the sum of the partial currents for all electrochemical
reactions in the reactor and a current balance must be done on the
electrode su rface
(87)
where itot is the total current flowing through the electrode.
The result of these balances on mass and current means that an

electrochemical reactor has a unique property, namely that an overall
reaction rate may be directly imposed and the individual reactions will take
up rates to satisfy this constraint. This is utilised by operating the reactor
at a constant imposed current and thus the current and mass balances
must be solved simultaneously in order to model the system. This is
analogous to mUlti-component distillation under constant boil up
conditions.
In distillation the partial pressure of a species dictates the rate of mass

transport of the species between trays and the total pressure within the
column is the sum of the partial pressures of the individual species. The
column pressure is then
m
_ ~
LJ a J. . x J.. p~at
J
(88)
j=1
where Ptot = total pressure in the column

Pj = partial pressure of species j
lXj = activity coefficient for species j
115
Xj = mole fraction of species j in the gas
PJ~at = sat urate d vapour pressure of species j
The saturated vapour pressure of a component may be calculated from

the Antoine equation which relates vapour pressure to temperature
= A.exp(- B ) (89)
C+T
where A, Band C are Antoine coefficients and T is the temperature of the

gas. This is analogous to the Tafel equation relating current density to
polarisation potential in the electrochemical system.
The total pressure is therefore given
= ~ a· ·x· ·A .. exp(- C.B- T J

£.J J J J
j
(90)
J
Under constant boil up conditions a constant pressure is imposed upon

the column and the pressure and mass balances must be solved
simultaneously.
In the electrochemical reactor the partial current density for each species
at an electrode may be related to activation (electrode potential) and mass
transport by the polarisation curve. The equation for this curve has been
derived in the previous section and the relationship for a reaction
exhibiting Tafel behaviour is described by equation (72).
When a current is imposed on a cell, each electrode takes up an

appropriate potential, even to the point of breaking down an apparently
neutral substance (eg solvent) in order to achieve the imposed current. In
the distillation analogy the temperature rises to maintain the required
pressure. The electrode potential will be the same for all reactions taking
place at the electrode surface. This can be calculated by simultaneously
solving equations (72) and (87) for each electrode and species.
116
The solution of these simultaneous equations can become complex for
multiple reactions but can be simplified for the aluminium reduction cell by
making the following assumptions:
1. The only electrochemical reaction at the cathode is the reduction of

aluminium.
2. During normal operation the only electrochemical reaction at the

anode is the oxidation of oxide. The partial current for the oxide
reaction is therefore the total current.
3. During the transition from normal operation to anode effect both

oxygen and fluoride ions are oxidised at the anode. However, for
anode effect to have initiated, the concentration of alumina in the
electrolyte must have fallen to a low level and it will be assumed that
the oxide reaction has become mass transfer limited. The equation
for the partial current of the oxide reaction, 10 , then reduces to
(91 )
and can be easily calculated. The partial current for the anode effect
reaction, the oxidation of fluoride ions, is then the remainder of the
total current through the cell
(92)
and the polarisation voltage can then be easily calculated from

equation (72).
4.6.3 Anode effect

Although there has been much research into the causes and conditions
for the occurrence of anode effect, most of the experiments have been
carried out in laboratory cells and have not attempted to quantify a model
for the increase in voltage. A good summary of the laboratory research
and the reaction model proposed to interpret the anode effect in laboratory
cells under potentiodynamic conditions is given by Calandra et aH14].
117
As has been stated previously, the anode effect in an aluminium reduction
cell results from a change in the electrochemical reactions occurring at the
anode. During normal operation, electrolysis depletes the concentration of
oxygen ions in the electrolyte until the concentration reaches a level
where the reaction becomes mass transfer limited and the reaction is
unable to proceed at the rate imposed by the current. As the cells are
operated at a constant current alternative ions will be oxidised at the
anode to maintain the overall rate of reaction. In the cryolite melt it is
fluorine ions that are discharged.
The fluorine discharged reacts with the carbon anode to form various
carbon fluorides, ie CF4 , C2FS etc. Unlike the carbon dioxide formed under
normal operating conditions, fluorine is very effective at preventing the
wetting of the anode surface by the electrolyte. The fluorocarbon bubbles
'stick' to the anode surface and coalesce into an insulating gas film. This
film spreads rapidly over the anode surface resulting in a high ohmic
resistance across the anode-electrolyte interface.
The fluorocarbon gas film is tenacious, and whilst it covers the surface,
normal anode reactions cannot be resumed, hence the requirement of
mechanical agitation of the anodes in anode effect quenching routines.
A typical plot of voltage against current density for an electrochemical cell

is shown in figure 4.10. As the current density increases beyond the
limiting current density of the first reaction, a second reaction, eg the
discharge of fluorine, will occur to achieve the imposed current. As the
concentration of the species involved in the first reaction falls, so the
limiting current density for the first reaction decreases.
The decomposition potential for the oxidation of fluorine ions is around

1.87 V compared with 1.2 V for the discharge of oxygen and the fluoride
reaction has been assumed to exhibit Tafel behaviour. As the
concentration of fluoride ions in the electrolyte is so great and anode
effects are relatively infrequent and short lived, the fluorine ion
concentration can be assumed constant and its effect upon the reaction
can be ignored.
118
..5!2
~
c::
~
-
0
0..
c::
~
~
'-
-
a
Ql
0')
·s
.(J)
"5
~
~
I c::
a:l ::J
0
0
~
()
W 0
-.-ttS- ~
'+-
0
~
«
+-'
c:
Q)
+-'
0
-
::J
.~
~
a.. Ql
"-
.E
~::J
«
I
-c::
()
.Ql
OJ <C .t:::
E:-
_--I
• ...:::::- I
I
I
E:-
___ I
• ...:::::- I
I
I
-Q)
0
0..I
.0
'-
I I
I
I (J)
I
c::
Q)
-
a
c::
~
"-
::J
()
a
.,.-.
~
~
:J
~
The cell voltage during anode effect in the cells at Anglesey Aluminium is
around 30 V, far in excess of the decomposition voltage associated with
the discharge of fluorine. This is due to the insulating effect of the
fluorocarbon gas film. As the gas film spreads across the anode, so less
of the anode area is able to conduct the current into the electrolyte and
the effective current density increases. An increase in current density
increases the rate and polarization voltage for the fluoride reaction and so
the film continue to grow until a point is reached where the ohmic
resistance of film is less than the resistance due to the reaction. The
current then arcs across the gas film and it is assumed that under these
conditions a conductive gas plasma is formed. Electrolysis will then take
place at the gas/electrolyte interface.
In order to fully model the anode effect, it is therefore necessary to

calculate the surface coverage of the anodes by the fluorocarbon film.
Assuming that carbon tetrafluoride, CF 4 , is the only fluorocarbon formed
and that the electrolytic discharge of fluorine is the rate limiting step in the
formation of carbon tetraflouride, the rate of formation of the gas film is
l
--·CEF·l/J moles S-1 (93)
4·z F ·F
where l = cell current density

CE F = current efficiency for the discharge of fluorine
l/J = a fittable parameter representing the leakage of gas
from the gas film
The surface coverage of the anode is the ratio of the bubble area to the
effective anode area. Assuming that the bubble is of uniform thickness
and that the gas will obey the ideal gas law, the rate of change of anode
coverage during anode effect becomes
de 22400· Tb __1 _ . CE .l/J (94)

F
dt 298 . f 4 . ZF . F
where e = surface coverage

T b = electrolyte temperature
f = bubble thickness
119
The partial current density for each reaction at the anode during anode
effect then becomes
(95)
1-8 (96)
When the anode surface is completely covered, it will be assumed in this

work that fluorine will be discharged at the bubble/electrolyte interface and
that the voltage drop across the interface will be the sum of the voltage
drop across the gas film and the polarization voltage associated with the
fluorine reaction.
(97)
4.7 Cell Voltage, Energy Consumption and Cell Resistance
The cell voltage can be used to calculate the energy consumption of the
cell by Ohm's law
Q cell = I.Vcell (98)
where Qcell = power consumed by the cell W

I = current passed through the cell A
V cell = electrical potential across the cell V
The cell voltage is not simply the sum of the electrode potentials but must
include the ohmic drop across electrolyte and the cell structure. The total
voltage drop across the cell is therefore
V cell = Eanode + Ecathode + Velectrolyte + V structure (99)
where Eanode' Ecathode = polarisation voltage at anode and

cathode respectively
Velectrolyte = voltage drop across the electrolyte
V structure = voltage drop across cell structure
120
The polarisation voltages and voltage drop across the electrolyte have
been considered in detail above. The voltage drop across the cell
structure is due to the resistance of the materials used in the cell
construction and includes terms for the anodes and cathode, anode and
cathode bus, anode riser and cathode flex. The resistance of each section
was provided by Anglesey Aluminium and the data that will be used in this
work is tabulated in table 4.2.
The voltage measured in practice is the cell voltage and includes all the
above components. The cell resistance is calculated from this and can be
used to infer conditions such as the alumina concentration in the
electrolyte.
Reell (100)
4.8 Heat Balance
Models to determine accurate temperature profiles and the heat flow

within cells have been developed but the number of calculations required
for solution make them unsuitable for simulation of a cells dynamic
behaviour. A good approximation may be achieved by dividing the cell into
a number of larger zones each of which is assumed to be well mixed and
therefore at a uniform temperature. The schematic model of the cell and
heat flows is shown in figure 4.11. The cell is divided into 8 coarse
components: (A) the aluminium pad, (8) the molten electrolyte, (C) the
anode, (D) the cathode, (E) the frozen ledge in the bath, (F) the frozen
ledge in the metal, (G) the cell sidewalls and (H) the electrolyte crust.
In spite of thermal disturbances to the cell, the heat balance of the cell
over a long period follows equation (101) for the steady state treatment.
(101 )
where Q res is the average power input to the cell, Q R the net heat of
reaction, Q H the average heat required for the heating and dissolution of
raw materials and Qout the average total heat dispersed from the cell.
However, for the simulation of the dynamic behaviour of a cell, the heat
121
Component Resistance Q
Anode rod clamp 1.044x10-5

Anode stub 1.458x10-5
Anode bus 4.1x10- 7
Anode riser 3.8x10- 7
Cathode flex 2.0x10- 7
Cathode bus 1.24x10-6
Table 4.2 External Resistances

Sidewalls
Freeze
Electrolyte Tb
A Aluminium Pad Tm
Qc
Figure 4. 11 Schematic Diagram of Heat Flow in an Aluminium Cell

which changes momentarily must be considered. Its treatment IS
described precisely in the following sections.
4.9 Electrolysis Heat Balance
The output from this block is the net heat of reaction due to electrolysis.
The net heat of reaction is given
n
- ~ r,I ·~HRT·
L..J . .I
(102)
i=1
where = net heat of reaction due to electrolysis W

rj = rate of reaction of species i, mol S-1
~HR,T,j = heat of reaction based on species i at
temperature T, J mol- 1
In this work four reactions have been defined as the main reactions
occurring in the aluminium reduction cell. The primary reaction during
normal operation is given
[8]
Some of the aluminium is reoxidised by
During anode effect the reaction is assumed to be
AIF3(soln) + C(s) ~ AI(I) + 0.75 CF4(g) [10]
and loss of current efficiency is assumed to occur by the reverse reaction.
The net heat of reaction is then
122
2 4
+ -3 . rco 2dlSS
. . L1H 2 - -3 . rCF diss . L1H 3
4 (103)
The enthalpy of reaction at any temperature may be determined from the

heats of formation of the individual species involved in the reaction. Thus,
for the reaction
aA + bB ~ cC + dD
the heat of reaction is given
(104)
where H~ (i) is the heat of formation of species i at temperature T and can

be determined from
H~ (i) = H~~ (i) + JT CPi . dT

T~
+ ~ (105)
where CPi = specific heat capacity of species i

T ¢ = reference temperature K
A; = net latent heat for any changes in state of species i
between T and T ¢
Over a narrow temperature range such as that encountered under

standard operating conditions, the specific heat of a species may be
assumed constant and this reduces to
(106)
where T a is the average operating temperature and the heat of formation

calculated at this temperature. For reaction [8] this gives
123
The heats of reactions for [9] and [10] are similarly calculated.
4.10 Heat Losses
In this work heat losses from the cell are defined to be the heat transferred
from the molten electrolyte and aluminium pad to the surrounding. In
figure 4.11 the losses are shown as the heat transferred from the
electrolyte to the anodes, OA' from the electrolyte to the frozen sidewalls,
Obf, from the electrolyte to the crust, 0bc, from the exposed bath to the
cavity beneath the shields, Oe' from electrolyte through the centre trench,
Oct, from the metal pad to the cathode block, Oc and from the molten pad
to the frozen sidewalls, 0mf' In this work it has been assumed that the
resistance to heat transfer between metal pad and electrolyte IS
insignificant with the result that heat transfer between these zones IS
instantaneous and the temperatures of these zones are equal.
Heat may be lost from the cell by convection and, in the case of exposed
bath, by radiation. Although the heat transfer coefficient for convection
may be calculated from the Sieder-Tate equation for variations in physical
properties and temperature this method is more suited for local heat
transfer coefficients. As physical properties change little in the range of
operating conditions and the velocity of the electrolyte and metal pad are
largely affected by the anode configuration and cell design, a
representative coefficient will be taken for each section and fixed for the
duration of the simulation.
4. 10. 1 Heat transferred to the frozen sidewalls

The heat transferred to the frozen sidewalls from the electrolyte and metal
pad is by convection only. In both cases the temperature at the ledge
interface is assumed to be the liquidus temperature of the electrolyte. In
the case of the metal ledge boundary this may be assumed due to the
existence of a thin film of electrolyte between the metal pad and the frozen
124
ledge. This film circulates with the bulk of the electrolyte and will be
assumed to have the same composition as the bulk electrolyte. The heat
transferred to the ledge is then
(108)
(109)
where h = heat transfer coefficient

A = area for heat transfer
T b = bulk temperature of electrolyte and metal pad
T 1iq = liquidus temperature of the electrolyte
suffix
bf refers to bath/ledge interface
mf refers to metal pad/ledge interface
4.10.2 Heat transferred from metal pad to cathode block
The heat transferred to the cathode block is by convection from the bulk of
the metal pad to the surface of the carbon cathode. The heat transferred
from the metal pad is then
(110)
where T Cs = the temperature at the cathode surface

suffix mc refers to the cathode/metal pad boundary
The generation of Joule heat, due to the electrical resistance of the

cathode block, and the inclusion of highly conductive metal collector bars
within the cathode block require a heat balance be performed upon the
cathode for the calculation of T Cs. As the thermal dynamics of the cell are
far quicker than those of the cathode, the cathode block will be divided
into two unequal zones as shown in figure 4.12. This allows the surface
temperature to react quickly to changes in the electrolyte temperature
whilst maintaining the overall heat losses and reducing the number of
calculations required in simulation. The heat balance for each zone is then
125
Zone Cs
Qcoll
Cathode Block
Alumina Zone C2
Firebrick
Qbott
Figure 4. 12 Heat balance across the cathode

(111 )
_ 12 kc x C2 + K'cXCS (T _ T )
A A Cs C2
C C
- UcaPcxc2 (Tc2 - Too)

- UCOIl Acoli (Tc2 - Too) - UbottAbott (Tc2 - Too) (112)
where Pc = density of cathode block

x = thickness of zone
CPc = specific heat capacity of cathode block
Ac = area for heat transfer through cathode block
T = temperature
kc = electrical conductivity of cathode block
K'c = thermal conductivity of cathode block
U = overall heat transfer coefficient
Pc = cathode block perimeter
Cs refers to cathode surface
ca refers to heat transfer from cathode to surroundings
C2 refers to second zone in cathode
coli refers to collector bar
bott refers to bottom of cell
00 refers to cell surroundings
4. 10.3 Heat transferred from electrolyte to cavity

In this work the heat transferred to the surface of the electrolyte has been
defined in three distinct ways, as heat transferred to the crust, heat
transferred from exposed bath and heat transferred through the centre
trench.
For simplicity, the surface crust will be assumed to be always in contact

with the electrolyte and will be treated in the same manner as the frozen
sidewall. This necessitates that a mass balance also be performed upon
the crust allowing freezing or melting into the electrolyte
126
(113)
dM crust Q be - Ucrust Acrust (T1iq - Tcav )

(114)
dt (~ +CPe (Tb - T1iq ))
where At = latent heat of fusion of electrolyte

CPe = specific heat capacity of electrolyte
suffix
bc refers to heat transfer from electrolyte to crust
cav refers to cavity beneath shields
For a centre break cell, heat transfer in the centre trench is alternately
from exposed bath after breaker action and by convection to an insulating
layer of alumina after a feed has been dumped onto the surface. As the
alumina has good insulating properties and does not remain on the
surface long enough to form a complete agglomeration with the electrolyte
by capillary action there is little formation of a frozen layer as there is with
the crust. The heat lost through the centre trench will then be by
convection and radiation for exposed bath and predominately by
convection and conduction for alumina cover. For simplicity, the heat
transfer through the alumina cover will be assumed to be at steady state
and represented by an overall heat transfer coefficient including the heat
transfer coefficients at both surfaces and the thermal conductivity of the
alumina layer. Therefore, for exposed bath
(115)
and for alumina cover
(116)
where hete = heat transfer coefficient from exposed bath in

centre trench to cavity
Eet = emmissivity from exposed bath in centre trench
to cavity
a = Stefan Boltzmann constant
Act = area of centre trench
127
Uct = overall heat transfer coefficient in trench
for alumina cover
The overall heat transfer coefficient is given for steady state flow of heat
through a composite wall as
1
U (117)
where U = overall heat transfer coefficient based upon the

inner surface area
h = heat transfer coefficient at wall surface
x = thickness of a layer
k = thermal conductivity of a layer
suffix
i refers to inner surface
j refers to jth layer of composite wall
o refers to outer surface
The overall heat transfer coefficient across the alumina cover in the centre
trench, where the area for heat transfer is assumed to be equal across
each surface, is then
1 1
(118)
where suffix
bt refers to heat transfer from bath to alumina
ac refers to alumina cover
tc refers to heat transfer from trench to cavity
Heat is also lost from exposed bath whilst the anode is removed from the
cell when a spent anode is being replaced. The operation of anode
removal lasts for several minutes and during this period the shield above
the anode must also be removed. Heat transfer is therefore to the air
surrounding the cell which may be at a significantly lower temperature
128
than the cavity beneath the shields. The heat transferred from the
electrolyte will therefore be
= hex ·A ex '("[b -Teo) + A ex . c~.~

~ ex
.("14
Ib
_T4)
co (119)
where the suffix ex refers to exposed bath.
4.10.4 Heat transferred from electrolyte to anodes
Heat losses to the anodes constitutes one of the major losses from the
electrolyte. Heat is transferred from the electrolyte to the anode by
convection and conducted away from anode/bath interface to the upper
surface where it is dispersed. The replacement of a spent carbon anode in
the cell with a new one at a temperature several hundreds of degrees
Celsius below that of the cell constitutes a massive thermal disturbance.
The initial effect is the formation of a frozen layer of electrolyte upon the
anode surface which effectively insulates the anode and reduces the rate
of heat loss from the bath. This layer initially grows, the latent heat of
fusion being given up in heating the anode, until either the anode has
reached a temperature such that the heat conducted away from the frozen
surface is less than that transferred from the bath and the layer melts, or
the frozen layer contacts the aluminium pad. In the latter case, the very
low liquidus temperature of aluminium prevents further freezing and heat
is conducted through the frozen layer and into the anode. The speed at
which the anode is heated is thus increased and anodes are frequently
'set low' to take advantage of this effect. As the anode warms and starts to
draw current, Joule heat is generated within the anode further warming it.
The complex nature of this problem requires that a heat balance must be
performed on each of the anodes and for this purpose each anode will be
divided into a number of unequal zones as shown in figure 4.13. As with
the cathode heat balance, the dynamics at the anode surfaces are
considered to be most important and the zones in these regions are
consequently the narrowest. Heat transfer is assumed to be uni-directional
through the anode with no heat losses from the sides.
There are three methods of heat transfer from the electrolyte to an anode
that must be considered. These are as follows:
129
Anode zone 2
Anode zone 1
Figure 4. 13 Heat balance across an anode

1) Convection directly to the anode surface: This is the case when
there is no frozen layer on the anode surface. The heat transferred from
the bath and into the anode is then given
(120)
where Q Ai = heat transferred from bath to anode i

QAs = heat transferred into surface layer of anode
hba = heat transfer coefficient from electrolyte to anode
surface
AA = anode area
T As = surface temperature of anode
2) Convection to frozen layer: When a frozen layer exists at the anode

surface and does not contact the metal pad, heat is convected from the
electrolyte to the freeze and conducted through the freeze to the surface
of the anode. If the heat convected into the boundary layer is greater than
that conducted to the anode then the layer melts and vice versa. As in the
model for heat transfer to the frozen sidewalls this model assumes that
the temperature at the electrolyte/freeze interface is the liquidus
temperature of the bath. The heat transferred from the electrolyte is then
(121 )
and the heat conducted through the frozen layer into the anode surface is
(122)
where kf = thermal conductivity of frozen electrolyte

Xaf = thickness of frozen layer at the anode surface
Any difference between the heats into and out of the boundary layer
results in freezing or melting of the layer. As with the heat balance upon
the crust the rate of freezing can be determined by a performing a heat
balance upon the boundary layer
130
(123)
which gives the rate of change of thickness of the frozen layer
dX af
dt (124)
dM af
where dt = rate of freezing of frozen layer at anode surface kg S-1
= rate of change of thickness of frozen layer on anode
surface m S-1
Pt = density of electrolyte at liquidus temperature kg m-3
3) Conduction through frozen layer: The low liquidus temperature,

about 660 °e, and high thermal conductivity of molten aluminium prevents
a frozen layer of aluminium from forming on any anode freeze that
contacts the metal pad. Growth of the frozen layer therefore ceases if the
thickness exceeds the interelectrode gap. Heat transfer to the anode is
then by conduction through the frozen layer:
= QAS -
~.A A .("[b - TAs ) (125)
Xaf
The rate of change in the temperature at the surface of the anodes is

determined by performing a heat balance upon each anode similar to that
performed upon the cathode. The anode is divided into three zones shown
in figure 4.13 and the heat balance for each zone is
(126)
131
- Kcov . xcov . (T - T )
A A2 A3 - Ustub · Astub . (TA2 - TcaJ (127)
A
C A dTA3 Kcov . xcov . (T - T ) - h. .( )

Pcovxcov Pcov cov dt = A A2 A3 cov Acov TA3 - Tcav
A
- c cov · Acov . (J . (T:3 - T!J (128)
where Ii = current through anode i

ccov = emmissivity of alumina cover on top of anode
suffix
a refers to anode
As refers to anode surface
cav refers to cavity beneath shields
cov refers to alumina cover on top of anode
stub refers to anode stub
A2, A3 refer to zones 2 and 3 in anode model
The total rate of heat transferred from the electrolyte to the block of
anodes is simply the sum of the individual rates. For n anodes this is
(129)
The total heat lost from the cell is the sum of the individual components:
(130)
4.11 Pot Heat Balance

As this work assumes that the resistance to heat transfer between the
electrolyte and metal pad is negligible the two layers are considered as a
single thermal mass. This combination will be referred to as the pot.
132
The overall heat balance on the pot is:
rate of accumulation - rate of electrical heat generation

of heat due to ohmic drop across pot
- total heat of reaction
- enthalpy in materials leaving pot
- heat required to raise feed materials

to temperature of pot
- heat of dissolution of materials added to the pot
- heat losses from the pot
The heat balance on the pot is taken relative to a reference temperature. If

this datum is chosen to be the temperature of the electrolyte then all the
terms for the enthalpy of materials leaving the pot become zero. These
include the enthalpy in aluminium tapped from the cell and the enthalpy of
the flue gases. The heat balance then becomes
(131 )
where Mpot = total mass of electrolyte and metal pad

= Mbath + Mpad
= net specific heat capacity of electrolyte and metal pad
Mbath . CPbath + Mpad . Cp AI
CPbath = specific heat capacity of electrolyte
CPAI = specific heat capacity of molten aluminium
ORes = heat due to the voltage drop across the electrolyte
OR = net heat of electrochemical and chemical reactions
0diss = heat of dissolution of feed materials
133
Qfeed = heat required to raise feed materials to the
temperature of the electrolyte
The heat due to the voltage drop across the electrolyte must be calculated
from the cell voltage excluding the voltage drop across the cell structure.
va = Vcell - V structure (132)
(133)
This is the only Joule heat that enters the electrolyte. The resistance of
the cell structure causes heat to be generated within the structure itself,
for example in the cathode block, and does not directly affect the molten
electrolyte.
4. 11. 1 Heat of dissolution

All materials dissolved in the bath consume heat proportional to the
amount of material added. The heats of dissolution for materials dissolved
in molten cryolite are not well documented but an average value for
alumina has been quoted by Thonstad[15] as 125 kJ mol- 1 . This value will
be used in this work as the heat of dissolution of the entire alumina feed,
inclusive of the added fines and sodium oxide. The heat of dissolution is
therefore given
Qdiss = Hdiss x (rdiss(suspn) + rdiss(sludge)) (134)
where Hdiss is the heat of dissolution of alumina.
The only other raw material added to the electrolyte in any quantity is
aluminium fluoride. As this is added infrequently and in relatively small
amounts compared to the alumina feed, 25 kg AIF3 per day as to 90 kg
AI 2 0 3 every 70 minutes, the relative disturbance to the heat balance is
very small. In this work, the aluminium fluoride is assumed to dissolve
instantly and the heat of dissolution is assumed negligible.
134
4. 11.2 Specific heat of feed materials
The feed materials added to the cell are much colder than the molten
electrolyte at about 200 0 e and heat is required to raise them to the
temperature of the melt, about 970 o e. The amount of heat required to
raise a substance from temperature T 1 to T2 is given
(135)
Although the specific heat capacity of feed materials may be assumed

constant over the range of operating temperature, there is a significant
variation between 200 and 970 o e. Tables of heat content are, however,
available for alumina and the heat required to raise the alumina feed from
200 0 e to temperature T may be found by interpolation
+ H1027 - H927 . (T - 927) - H200 (136)

1027-927
using JANAF data[16] gives
1
H927 - 103.24 kJ mol-
H 1027 - 116.08 kJ mol- 1
H200 - 16.41 kJ mol- 1
and
HA1203,T = 86.83 + 0.1284 ( T - 927 ) kJ mol- 1 (137)
As alumina is the most significant component of the feed, this relationship

will be used for the heat content of the entire feed inclusive of fines.
The heat loss due to addition of aluminium fluoride is much less significant
and for simplicity, the specific heat capacity of this component will be
assumed constant. The heat required to raise AIF3 additions to the
operating temperature, based upon data from Aylward and Findlay[17] is
HA1F3 , T = epAIF3 ( T - 200 ) = 75·(T-200) (138)
135
4.12 Frozen Sidewalls
Models for the precise estimation of the frozen sidewall profile have been
presented[18,19,20] but require the division of the cell into a number of
triangular sections. A mass and heat balance must be performed across
each section and the time required for these calculations makes these
models unsuitable for dynamic simulation.
The profile of the frozen sidewalls is of prime importance in the study of

the hydrodynamics of the cell. In this work, a simple hydrodynamic model
is assumed and the detailed calculation of freeze profile is unnecessary.
The frozen sidewall will be divided into two zones, that in contact with the
molten electrolyte and that in contact with the molten aluminium. Two
zones are required because of the difference in heat transfer coefficients
in the electrolyte and aluminium pad.
In each zone the sidewalls are assumed to be of uniform thickness and

the heat flow unidirectional.
The heat transferred from the molten bath to the freeze and from the cell
outer wall to the surroundings is assumed to be purely by convection.
Heat transferred through the freeze and composite wall is by conduction.
The heat flow through each section is therefore
qb -- hb ·A b .(Tb - T, ) (139)
qf - ~.Af .(T, - Tfw) (140)

xf
q1 - ~.A 1.(Tfw -Tw1 ) (141 )

X1
q2 - ~.A2 ·(Tw1 - Tw2 ) (142)

x2
q3 -- ~.A s .(Tw2 -T)

s
(143)
Xs
-- h s ·A s .(Ts - Ta ) (144)
qs
The primary area of interest is the heat flow into and out of the boundary
layer at the surface of the frozen sidewall as this controls the melting and
freezing of the frozen sidewalls. The resistance to heat transfer of the
136
freeze, composite sidewall and outer surface may be approximated to an
overall heat transfer coefficient U, by
(145)
where the heat transfer coefficient is based upon the internal area of the
freeze. The equations for heat transfer across the boundary layer then
become
In qb = hb Ab ( Tb - Td (146)
Out qt = U At ( TI - T b ) (147)
In addition to reducing the number of computations required, this

approximation overcomes the problem of initialising the temperatures at
each material interface.
If the boundary layer is assumed to be infinitely thin and at the liquidus

temperature of the molten electrolyte then any accumulation of heat will
cause melting of the sidewalls to the temperature of the bath. Performing
a heat balance across the boundary layer gives
(148)
Change of state from solid to liquid Heat in Heat out

+ enthalpy of change in temperature
The rate of change of state of the frozen sidewalls is then given
dm (149)
dt
and the rate of change of the thickness of the freeze
(150)
137
The cell wall in the metal pad region consists of a sloping carbon section
and two layers of firebrick. As the cell ages the profile of the carbon
section changes as shown in figure 4.14 with the slope in the metal pad
area becoming less steep. In this model, the wall thickness in the metal
pad area will be assumed to be of a uniform thickness as is more the case
in an older cell. The assumptions made in general in this model do not
favour the unpredictable behaviour of a cell that has been newly brought
on line and the assumption of a freeze profile similar to that of an
established cell is therefore justified.
The diffusion of species between the frozen ledge and the melt during
freezing has been modelled by Hashimoto[21] and requires the estimation
of a diffusion coefficient for each species and the velocity of solidification.
A further constant, related to the diffusivity of the melt, must also be fitted
and estimates by Hashimoto range between 0.1 and 0.00001 for
commercial cells. The complexity of this model and the difficulty in
obtaining meaningful values for these parameters make it unattractive.
Chemical analysis of the composition of the frozen layer in spent cells at
Anglesey suggest that the composition of the freeze remains
approximately constant throughout the life of the cell, with weight ratios of
1.36 to 1.41 being obtained irrespective of the operating ratio in the bath.
This allows a much simpler model to be developed for the migration of
species between molten bath and frozen sidewall.
If the frozen sidewall is assumed to be of constant composition with a

weight percent of alumina of At, calcium fluoride of Ct and a weight ratio of
Rt then the mass of individual species transferred between the melt and
the ledge during freezing can be obtained
dM A , At dM
-- - . - -f (151 )
dt 100 dt
dM c, C dM
-- - f. - -f (152)
dt 100 dt
dM A'F3,' (100-A f -C f ) dM f
- .-- (153)
dt 100·(R f -1) dt
dM NaF,I dM f dM A, dM c, dM A'F3,1
-- (154)
dt dt dt dt dt
138
Frozen Sidewall
/ / / / Electrolyte / / / / /
/ / / / / / / / /
Aluminium Pad
:I Cathode
New Cell Old Cell
Figure 4. 14 Cell Sidewall Profiles in Old and New Cells

dM f
where dt = rate of freezing of bath at sidewall
dM.
dt If = rate of migration of species i between bath and ledge
4. 12. 1 Heat transfer coefficients
The heat transfer coefficients between the molten bath and the frozen
ledge and between the aluminium pad and the frozen ledge will vary
throughout the cell dependent upon the geometry and local conditions.
The object of this work is not to calculate a detailed freeze profile and the
hydrodynamic and temperature models already presented here would not
be sufficient for the task. Here, average heat transfer coefficients will be
assumed, with constant values for the bath and metal pad sections. This
will allow an average thickness of freeze and the total heat lost through
the sidewalls to be calculated in each section.
Both Arita and Haupin[22] have presented average values for the heat
transfer coefficients in bath and metal regions and these are shown in
table 4.3. The effective value for the metal zone presented by Haupin is
calculated by assuming a constant film of molten electrolyte to exist
between the ledge and the metal pad. Using these values as a starting
point, the values used in simulation may be tuned for individual cells to
give a stable ledge profile.
4.13 Anode Change

Changing an anode in a cell represents a large disturbance to both the
heat and mass balances. During its removal, alumina from the crust and
frozen sidewall are broken into the melt and a large area of bath is
exposed while the anode is removed. The spent anode is replaced by a
cold anode at ambient temperature causing freezing of the electrolyte on
the anode surface which disrupts the current distribution through the bank
of anodes.
139
source bath metal effective
Haupin 3.7x10-2 1.5x10- 1 1.1 x1 0- 1

Arita 2.0x10-2 4.0x10-2
(side break cell)
Table 4.3 Heat Transfer Coefficients in Aluminium Reduction Cells

4. 13. 1 Heat losses
The heat lost from the electrolyte during an anode change and whilst the
anode is warming may be divided into three stages.
1) Heat lost from exposed bath whilst the anode is removed from the
cell
2) Heat transferred through a frozen layer of bath to the cold anode
3) Direct warming of the anode once the frozen layer has melted
1) When an anode is removed from the cell, a large area of bath is

exposed while the new anode is manoeuvred into position. During this
time, heat will be lost from the electrolyte by radiation and convection.
QIOSS - heat lost due to forced convection +

heat lost due to radiation
where subscript b refers to bath

s refers to surrounding
hole refers to exposed area
A = area
T - temperature K
e - emmissivity chosen to include view factor
a = Stefan Boltzmann constant
The heat losses at this stage will last for the time it takes to replace an
anode in the cell. At Anglesey, this takes approximately four minutes.
2) When a cold anode is lowered into the cell, the heat lost due to direct
radiation and convection to the surroundings is immediately stopped. Heat
is now lost by convection to the surface of the cold anode. This causes a
layer of frozen electrolyte to form at the anode surface and the heat losses
will be treated in the same way as for the frozen sidewalls.
(156)
and the thickness of the frozen layer will be given by
140
dX f
dt (157)
where qanode = rate of heat transfer through the anode

Pt = density of frozen layer
Tb = Temperature of the electrolyte
T, = Liquidus temperature
A, = Latent heat of fusion
Cp = Specific heat capacity of the electrolyte
When the thickness of the frozen layer becomes greater than the anode to
cathode distance, the frozen surface will come into contact with the metal
pad. The heat transfer coefficient in this section is much greater than that
for the electrolyte and the heat transferred to the freeze will become
(158)
As the melting point of aluminium IS much lower than that of the

electrolyte, the frozen layer is unable to become thicker and the heat
transferred from the aluminium pad will warm the anode much more
rapidly than the heat transferred from the electrolyte. For this reason, new
anodes are usually 'set low' for the first day so as to allow the anode
surface to be closer to the metal pad thus increasing the rate at which the
anode can attain normal operating temperature.
Unlike the sidewalls, which are assumed to remain at a reasonably steady

state temperature, the anode is subject to a large increase in temperature
as it warms both by heat conducted through the freeze from the electrolyte
and by the passage of electrical current through the anode. The full heat
balance detailed in section 4.10.8 must therefore be performed on the
anode.
4. 13.2 Resistance of anode block

Initially, the frozen layer of electrolyte on the surface of the cold anode
prevents the flow of current through the anode. As the anode warms and
the frozen layer melts the anode is able to draw current. In order to model
this effect fully, it would be necessary to consider the current distribution
141
and resistance throughout the anode and freeze and this is beyond the
scope of this work. The model may be simplified by assuming a profile for
the surface coverage of the anode by an insulating layer. The resistance
of the anode can be modelled as two resistances in series, R1 relating to
the resistance across the anode and R2 relating to the resistance of the
frozen layer and the lower conductivity of a cold anode. The total
resistance of the anode is then
Ranode = R1 + R2 (159)
Assuming that the resistivity of the frozen layer is much greater than that
of the molten electrolyte in the same region gives
- PI
(160)
f·A
where PI = effective resistance of frozen layer per unit area

A = Anode surface area
f = fraction of anode area unable to conduct current
The fraction of the anode surface area unable to conduct current can be
related to time by an exponential function
f = a·e bt (161 )
where the pre exponential and exponential factors, a and b, are chosen to
match the current distribution profile for a cold anode obtained on working
cells.
The total resistance of a block of n anodes connected in parallel is given
1 n 1
= L-
i=1 Ri
(162)
and the current drawn by the ith anode in the block
I . Rblock (163)
- tot R.
I
142
The total area for current flow through the cell is the sum of the effective
areas through each anode
n
A Anodes - ~ 1,I ·A.I
L..J
i=1
(164)
giving an overall current density for the cell of
Itot
AAnodes (165)
In the cells at Anglesey Aluminium, it is known that a new anode in a cell

takes approximately 32 hours to draw full current. A short program solving
the heat and mass balances on an anode in a bath of constant
temperature allows the values of a and b to be fitted to give the desired
current profile. In this work, the values of a=1.7x1 0- 4 and b=6.69x10- 5 S-1
will be used, giving the profile shown in figure 4.15.
4.14 Phyico-Chemical Relationships
Many of the parameters in the model are determined from empirical

relationships relating properties of the system to conditions within the cell.
Empirical relationships for such properties as electrolyte density are
continually being published as further research is carried out on aluminium
smelting and this is an area of the model that may be subject to relatively
frequent change as improved and more applicable relationships become
available. Some of the relationships used in this model are summarised
here.
4. 14. 1 Alumina solubility

The following relationship is taken from KACC Technical ManuaH11] and
will be used in this work.
143
4 9000
3.5 8000
- ~ 7000 ;(
-"g
-...
E
(.)
CI)
>-
3 ~
6000
CI)
as
.!! 2.5
cCI) .c
N
5000 g'
...
0
..... 2 ...0
.c
.....0 4000 i
rn 0
~ 1.5 ~
c 3000 C
~
(.) ......
CI)
.c 1
t- 2000 8
0.5 1000
t , ! ! ! I I I I I I I I I I ,
0
I I
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Time since anode change (hours)
Figure 4. 15 Current Drawn by a Cold Anode

4. 14.2 Aluminium solubility
The solubility of aluminium in the electrolyte is quoted by Lillebuen and

Melierud[10] as
C~I = (0.0268 -Rb - 0.6755 + 10-3 - Tb -1.9 X 10-3 - %CaF2
-12.5))-1~~
3
-4.3x10- -(%AI2 0 3 g cm- 3 (167)
and the solubility of carbon dioxide
(0.03027 - 6.985 x 10-4 -%AI 0 )

- - , - - - - - - - - - - , - -2 -3- -1 0-3 g cm-3 (168)
(1- 0.08989 -%AI 2 0 3 )
4.14.3 Density
The density of the molten electrolyte will be calculated from a relationship

derived by Kvande and R0 rvik[23]
Pb = 2.64 - 8 x 10-4 -Tb + 0.18 -Rb

- 8 x 10-3 - %A1 2 0 3 + 5 x 10-3 - %CaF2 kg m- 3 (169)
The density of the molten aluminium is calculated from the following

relationship[11]
PAl = 2.561-2.72x10-4-T kg m-3 (170)
In both equations, the temperature is in degrees centigrade.
4. 14.4 Liquidus temperature

The relationship for liquidus temperature was reported by Bullard[24] after
work by Foster and Hollingshead.
Tliq - 733.6+392.5-R b -164.9-R~ +15.35-R~

(171 )
- 3.375· o/OCaF2 - 5.4 - %A1 20 3
144
4. 14.5 Viscosity of the electrolyte
A mathematical model for the viscosity of aluminium reduction cell

electrolytes has been fitted to empirical data by Hertzberg et aj[25J. The
model will be used in this work.
In lib ~ C~~} In liN" + x{ dy- x)exp[ -by {dy- x)']ex+ :~}n Ii.
+ Y( d x - Y)exp[ -b x(d x - y)2 Jexp[ -cxx]ln,uy
+ x2 :100 exp[ -e 1 (25 - X)2 ]exp[ - e2y2]In,ux (172)
where
100X A1F
X= 3
In,uy = 0.01502 + 47.22t by = 8.515x10-4
XNaF + XA1F 3
Y = wt% AI 2 0 3 In,ux = 1.188 + 5300t c y =0.6111

t=-.!- 1 d y = 37.8gexp(619.3t) b x = 0.002657
T 273.14
In ,uNaF = 0.37758 - 3184t d x = 30.12exp(-508.9t) e 1 = 0.05
In,ux = 0.001934 + 1O. 99t Cx = 0.04051exp(1948t) e 2 = 0.02253
Viscosity is predicted in m Pa s. Temperature is in degrees centigrade.
4.15 Summary
This chapter has described the formulation of the physico-chemical model

of an aluminium cell that will predict how the cell will react to such actions
as movement in the anode beam position and the breaking of the crust
and addition of feed materials. The next chapter describes the modelling
of the existing control system used at Anglesey Aluminium and discusses
the implementation and simulation of alternative control strategies.
References
1 Blake, S. R., "A Model for the Variation of Alumina Concentration in

Industrial Hal/-Herault Cel/s", MSc Thesis, Department of Chemical
2 Entner, P, Scmidt-Hatting, W., Mitrovic, Z., Gruber, U., Rufer, D. and

145
3 Johnston, A. R., "Alumina Crusting and Dissolution in Molten
Electrolyte", AIME Light Metals, 1981, pp373

5 Asbj0rnsen, O. A., Andersen, J. A., "Kinetics and Transport Processes in

the Dissolution of Aluminium Oxide in Cryolite Melts", AIME Light
Metals 1977, pp137-152
6 Haupin, W. E., "Mathematical Model of Fluoride Evolution from Hall-

Trondheim, 1985
7 Sidorov, L.N., Kolosov, E.N. and Shol'ts, V.B., Russian Journal of

Physical Chemistry, 42, pp1382-1386, 1968
8 Robl, R. F., Haupin, W. E., and Sharma, D., "Estimation of Current

9 Evans, J.W., Zundelevich, Y. and Sharma, D., Met. Trans, vol 12B,
1981
10 Lillebuen, B., and Mellerud, Th., "Current Efficiency and Alumina


12 Plimley, R. E, "Process Modelling and Optimisation", Electrochemical

Process Engineering Course, University of Newcastle upon Tyne, 1984
146
13 Zuca, S., Herdlicka, C. and Terzi, M.,"On Porosity-Overvoltage
Correlation for Carbon Anodes in Cryolite-Alumina Melts",
Electrochimica Acta, vol 25, February 1980, pp211-216
14 Calandra, A.J., Castellano, C.E., Ferro, C.M. and Cobo, 0.,

"Experimental and Theorectical Analysis of the Anode Effect in
Industrial Cells", AIME Light Metals, 1982, pp345-358
15 Thonstad, J., Nordmo, F. and Paulsen, J. B., "Dissolution of Alumina in

Molten Cryolite", Metallurgical Transactions, vol 3, February 1972,
pp403-407
16 JANAF Thermochemical Tables, NRDS, National Bureau of Standards

37
17 Aylward, G. H. and Findlay T. J. V., SI Chemical Data, 2nd Edition, The

Jacaranda Press, 1974
18 Arita, Y., Urata., N. and Ikeuchi, H., "Estimation of Frozen Bath Shape
in an Aluminum Reduction Cell by Computer Simulation", AIME Light
Metals, 1978, pp59

Metals, 1978, pp73
20 Peacey, J. G. and Medlin, G. W., "Cell Sidewall Studies at Noranda

21 Hashimoto, T. and Ikeuchi, H., "Computer Simulation of Dynamic

pp273
22 Haupin, W. E., "Calculating Thickness of Containing Walls Frozen from

147
23 Kvande, H. and RC:Hvik, H., "The Influence of Bath Density in Aluminium
Electrolysis", AIME Light Metals, 1985, pp671-678
24 Bullard, G., KACC Research Report, 2nd quarter, 1984
25 Hertzberg, T., T0rklep, K. and 0ye, H.A., "Viscosity of Molten NaF-AIF3 -

AI2 0 3 -CaF2 Mixtures", AIME Light Metals, 1980, pp159-170
148
5.0 The Control System
The overall control aim of an aluminium reduction cell is to maintain good
current and energy efficiencies. The system is, however, highly complex
and control involves the manipulation of several process variables at
regular and irregular intervals.
There are three fundamental problems which the process presents to all
process control systems. These are:
1. A multiplicity of relatively small but interdependent production units

each of which must be individually controlled. Typically a potline may
comprise over 100 reactors and a plant a number of potlines.
2. Lack of equipment to continuously and accurately add raw materials to

and remove product from the process, although the process itself is
continuous.
3. Lack of instrumentation and sensing devices to rapidly and accurately

measure and control process variables.
These limitations have resulted in the reliance of the control system upon
considerable manual intervention based upon the judgement of
experienced cell operators. The use of digital computers and better
mechanisation has, however, improved controllability of several but not all
of the process variables.
The main tasks that must be performed in order to control the cell are as
follows:-
1. Control of alumina feed rate for a desired anode effect frequency

2. Maintaining a desired operating voltage and temperature in the cell
3. Removal of liquid aluminium product
4. Maintaining a desired electrolyte composition
5. Maintaining a desired inter electrode gap
6. Maintaining an uniform current distribution
7. Regulating the thickness and shape of frozen sidewalls
149
The way in which these tasks are implemented constitutes the control
strategy.
5.1 Alumina Concentration Control
The mass flow of materials into and out of the cell consists of the flow of
solid raw materials to the cell and the flow of reaction products from the
cell. Although the production of aluminium in the reduction zone is
continuous the addition of the major raw material, alumina, is batchwise
and it is this task that must take priority in the control scheme. If the level
of alumina in the electrolyte rises to a high level, a sludge is deposited at
the cathode causing poor operation of the cell. Ultimately, overfeeding
may result in the density of the electrolyte exceeding that of the liquid
aluminium and the two layers will invert with disastrous consequences.
Alternatively, if the alumina concentration is allowed to fall below a critical
level, anode effect occurs with the production of fluorinated carbon
compounds at the anode accompanied by a large increase in cell voltage.
There is, at present, no means of accurately determining the concentration

of alumina within the electrolyte on a continuous basis. The alumina feed
rate must therefore be determined by inference from other parameters,
notably voltage, and the calculated current efficiency of the cell.
5.2 Removal of Aluminium Product
The removal of aluminium product from the cell is also a batch operation
with material being siphoned from the metal pad at regular intervals. The
depth of the metal pad has an important effect upon the heat balance of
the cell and upon the current efficiency by affecting the pad stability. More
importantly though, the amount of aluminium removed gives an indication
of the rate of production of aluminium in the cell and is used to calculate
the overall current efficiency. This is used as a measure of the
performance of the cell over a long period of time and is used in tuning
other control variables.
The amount of metal removed from the cell is determined by the depth of
the aluminium pad within the cell. Both the depth of pad and the amount of
150
metal tapped are generally measured by a crude dip stick method. The
molten aluminium is siphoned off into a crucible. The amount tapped is
measured by level in the crucible and is calculated to leave a desired level
of metal pad depth in the cell. The recent introduction of electronic scales
to weigh the crucible has improved the accuracy of the tapping operation
and hence the calculated current efficiency but the control of metal pad
depth is still very crude.
5.3 Electrolyte Composition
It is also necessary to maintain the composition of the bath by

replacement of materials lost during anode effect and by bath
volatilisation. The bath composition has a major effect upon the physical
properties of the electrolyte and most importantly upon the electrolyte
conductivity and current efficiency. With the exception of alumina and
aluminium fluoride, the rate of change of concentration of particular
species is small and does not take a high priority in the short term. The
effect of aluminium fluoride concentration on the ratio NaF/AIF3 is,
however, more important and maintaining this ratio has a higher priority.
Again there is no method for the continuous monitoring of bath

composition but it may be determined by analysis of bath samples taken
weekly. Based upon this data over a period of time the average amount of
material lost may be calculated. This is replaced by the addition of
recovered exhaust materials to the alumina feed and by the addition of
bags of aluminium fluoride to the melt at periodic intervals.
5.4 Anode to Cathode Distance

As well as controlling the ohmic drop across the electrolyte the anode to
cathode distance affects the current efficiency and stability of the cell.
Below a critical value the current efficiency falls off rapidly but above this
level there appears to be little or no change with increasing interelectrode
gap. Theoretically there is an optimum gap for any composition and
temperature of electrolyte. There is no direct method of determining anode
cathode distance although some control may be conferred upon it through
overall cell voltage control.
151
The inter electrode gap may be altered by moving the anode bus bar
relative to the cathode. This is done by pneumatic motors that may be
activated either manually or automatically. The motors act upon the whole
anode block, not individual anodes. Any anodes that are badly positioned
relative to the block must therefore be reset manually.
5.5 Current Distribution
The maintainance of a uniform current distribution through the cell is

desirable as it benefits the stability and hence the efficiency of the cell. In
general, for a uniform distribution, the inter electrode gap for each
individual anode should be equal. Due to the electrohydromagnetics of the
cell and distortions in the cathode block, however, there tends to be a
'hump' in the middle of the metal pad. This requires that the anodes be set
at different heights relative to the anode bus bar according to their relative
position in the cell.
The current distribution through a cell may be measured manually by

measuring the voltage drop across each anode stub. This is done each
day and any anodes that are not pulling the expected current are marked
down for resetting. As well as detecting poorly set anodes, this method
can detect anodes that have short circuited to the cathode by 'spiking' or
anodes that have burnt off from the stubs.
5.6 Voltage and Temperature Control
For a cell operating at a constant current the cell voltage determines the
energy input and consequently temperature. Electrolyte conductivity is,
however, affected by temperature and so to some extent, temperature
affects cell voltage. Another effect of temperature is to alter the freeze
profile and consequently cell voltage. These two parameters therefore
show considerable interaction.
A low voltage drop across a cell is desirable in order to reduce the energy
required to produce the aluminium. A reduced ACD may, however, have
an adverse affect upon electrolysis by increasing the rate of back
reduction and reducing the current efficiency. At times when the market
152
price of aluminium is high it is preferable to maximise the production rate
over the energy required for production and vice versa.
Due to the highly corrosive nature of molten cryolite it is not possible to

continuously measure the temperature of a cell. Instead, a sheathed
thermocouple is used to measure the bath temperature daily and the
overall appearance of the cell is noted. If the temperature is consistently
high over a period of days and the cell lappearsl hot then the voltage set
point may be lowered. This can be continued until the cell lappearsl to be
operating satisfactorily. The temperature of the cell affects the current
efficiency by increasing the solubility of aluminium in the electrolyte but
more importantly it has a marked effect upon the frozen ledge profile.
5.6.1 Frozen sidewalls
The frozen sidewalls contain the molten electrolyte and aluminium and
protect the cell walls from their corrosive effects. They also regulate the
temperature of the cell by melting or freezing. A good freeze profile,
however, is necessary for good current distribution and efficiency. If the
freeze extends too far under the metal pad then the area for current flow is
reduced. If the freeze becomes too narrow then the stability of the metal
pad is reduced and ultimately the sidewalls become corroded.
The design of the cell structure and the materials used in its construction
very much determine the shape of the frozen ledges, but the profile is also
affected by the temperature and composition of the electrolyte. Although
the freeze profiles cannot be measured easily they may be controlled by
maintaining the electrolyte temperature and composition about the design
conditions. The correctness of the operating temperature is, however, only
reflected in the long term current and energy efficiency of the cell and it is
this information which is used to alter the voltage set point.
5.7 Manual Control

The simplest method of control is to assign an operator to control a cell or
group of cells. The operator manually adjusts the anode beam position to
control the cell voltage and feeds alumina and other feed materials into
the cell based upon observed conditions and control schedules. This type
153
of control relies greatly upon the judgement and commitment of the
operator and for a reasonable level of control it would be necessary to
have an operator assigned to only a small number of cells.
In most situations, some form of automatic control is preferable as it

requires less manpower and will generally provide tighter and more
consistent control.
5.8 Manual Control of Alumina Concentration

Alumina is fed manually by means of a mechanical crustbreaker. During
this operation crust falls into the bath along with some of the loose
alumina covering it. The alumina cover is subsequently replenished and a
new crust formed. The batch weight is very poorly controlled by this
method and the pot may be severely under or overfed. In order to ensure
that alumina concentration does not exceed a critical value, feeding may
be suspended to force periodic anode effects.
The termination of anode effect is not simply a matter of increasing the

alumina concentration within the electrolyte. In order for this to be effective
the carbon tetrafluoride gas bubbles which adhere to the anodes must be
dispersed. This is achieved manually by raising and lowering the anodes
and by the introduction of a Igreenl stick (wet hazel rod) beneath the
anodes. The bubbles from the burning stick cause enough turbulence in
the cell to break up the CF4 bubble. Streamers of aluminium are also
thrown up from the metal pad momentarily shorting the cell so causing the
electrolysis of fluoride to cease and the bubble to disperse.
5.9 Manual Control of Voltage

The cell voltage is the only parameter of an individual cell that may be
monitored accurately, continuously and cheaply. With the exception of the
voltage rise at anode effect which is controlled by the addition of alumina,
other fluctuations are controlled by altering the inter electrode gap.
Although anode movement may be activated manually by push buttons at
the cell side the control of these fluctuations would require an operator
dedicated to each cell. This an unfeasible task and manual voltage control
154
must concentrate upon achieving a long term average voltage by
maintaining an approximately constant ACO. Movement of the anode bar
is then only necessary when a setpoint change is required or during
tapping when the metal pad level is falling rapidly.
5.10 Manual Control Strategy
Central to the manual control strategy is the IIpotline surveyll. This survey
is conducted by the potline general foreman and involves the inspection of
individual pots on the line on a shift to shift basis. The specific attributes
inspected are:
flame colour
pot voltage and stability
carbon dusting in the bath
liquid levels.
All abnormalities are recorded and a severity factor ranging from 0.5-3.0 is
assigned to the cell based upon the overall condition of the cell.
Assignment of severity values is prescribed by the standard practice.
On completion of surveys upon all cells within the line, the individual cell
severity values and abnormal characteristics are tallied and summarised.
A complete record of the potline operating condition is thus available for
applying compensatory corrective action on individual cells. Review of
abnormal attributes and their previous history provides information for
formulating cause removal corrective action.
5.11 Automatic Control
In the late 1950's automation of some of the more tedious and laborious
hand operations on reduction cells began to take place. This initially
consisted of the mechanisation of crust breakers on prebake cells and
voltage control by means of mechanised anode movement controlled by
pneumatic motors activated by operators from the floor. This was soon
followed by analog electro-mechanical devices which broke crust, fed
alumina and controlled cell voltage on some predetermined schedule set
by potroom operating supervision. Automatic anode effect termination was
155
added as an additional task and the development of point feed cells
allowed smaller amounts of alumina to be fed at more frequent intervals
on a semi continuous basis.
Although automation had been applied to some aspects of process control

other tasks such as removing metal from the cell, control of liquid levels
and electrolyte composition remained unchanged. There was little
improvement in instrumentation or sensing devices for increasing the
amount of information immediately available about a cell's operating
condition.
In the mid 1960's digital computers were first incorporated into pot line
operation and process control. Although the environment is hostile to
these devices due mainly to the strong magnetic fields and lack of
cleanliness in the pot rooms this was overcome by the use of distributed
systems.
5.12 Distributed Control
In a distributed system a control unit is dedicated to each part of the

process with each unit linked back to a central computer for supervisory
control of the overall process. The specialization of the units and their
close proximity to that part of the process that they are controlling allows
them to track process activities and make control decisions much faster
and more accurately than a centralised system. As there is a high degree
of independence between the control units, the failure of an individual unit
need not prevent control of the process. The central computer now acts as
system controller for the control units rather than the unit controller itself.
5.13 The CELTROL System

The distributed control system used by Anglesey Aluminium is shown in
figure 5.1. The system consists of a microprocessor at each cell, a
communications processor for each room, a supervisory computer for the
plant and a backup/development computer. The microprocessor is known
as GELTROl and provides potroom personnel with alumina addition,
156
POTROOM
POTROOM ti5l Elm ti5l mml1 ti5l

AUDIO EmllOI Cml1 EmllOI Elm EmllOI
CELTROL CELTROL CELTROL CELTROL CELTROL
SUPERVISORY
COMPUTER
MODCOMP
ffl w ~
Figure 5. 1 CEL TROL Distributed Control System

voltage control, anode effect termination, stability control, line load control
and information feedback.
5. 13. 1 The eEL TROL unit: a microprocessor
Each GELTROl unit contains all the input/output functions and logic
necessary to control a reduction cell. The hardware allows for cell voltage
input and operator inputs through toggle switches, push buttons and
selector switches. These switch inputs are used for altering control
parameters at the cell, acknowledging alarm conditions and viewing
setpoints and some historical data. Outputs to the cell allow anode bus
movement and break and feed operations at the tap and duct ends of the
cell.
Due to this self contained nature, GELTROl requires very little

maintenance. If a unit should fail then only the control functions on that
cell are lost and may be restored by simply changing the panel, an
operation that only takes about 10 minutes. Similarly, new control routines
may be easily tested upon small groups of cells simply by replacing the
memory boards within the required panels.
The GELTROl units are connected to the room communication buffer by

means of a communication bus running the length of the room. Through
this bus, data is received or sent to the buffer and in turn to the
supervisory computer.
As well as directly improving control to individual cells, the high speed

data collection and information feed back facilities of GELTROl have
proved useful in making further operational improvements. GELTROl
samples cell resistance every 100 milliseconds and from the fluctuation in
l
these values over 1 minute periods calculates the Inoise level of the cell.
This parameter with historical information provides indications of abnormal
operating conditions in a cell.
5. 13.2 The communications buffer

As well as providing communication between the GELTROl and the
supervisory computer, the communications buffer reads the line amperage
and voltage signals. The line amperage is received from measuring
157
equipment in the rectifier and is sampled by buffer software every 100
milliseconds. These values are digitally filtered and broadcast to each
GELTROl every 500 milliseconds where it is used to calculate the cell
resistance.
location of the communications buffer in a more favourable environment

than that of the GELTROl units makes failure rare. If a failure does occur,
however, only the communication to that room of GELTROlls is lost. The
GELTROl units continue to perform their control functions upon individual
cells until the buffer is replaced.
5. 13.3 The supervisory computer
The supervisory computer is a mini computer located in a central

computer room at some distance from the pot lines and connected to each
of the communication buffers by a standard RS-232 link. It has five main
functions to perform:
1. detect and output alarm conditions on any cell

2. collect historical data for reports
3. allow line management to make control setpoint changes on any cell
4. perform line load control
5. make fine setpoint adjustments based upon historical data
Every thirty seconds the supervisory computer requests a status report

from each cell. From this report it determines whether any alarm
conditions exist and whether that condition should be output over the
audio system. The information from this report is also recorded on
magnetic disc and, if required, hard copy may be obtained on a printer.
Periodically the supervisory computer requests a memory dump from each

GELTROl unit. This contains historical data about the cells operation over
the period. This information may be compiled into reports on cell and plant
operation and used to make decisions on operational changes. The
supervisory computer itself has the ability to make routine decisions on
setpoint changes by utilising the historical data on a cell's operation.
The supervisory computer is also connected to terminals in each of the

line offices. Through these terminals, operators are able to view current
158
and historical operating data for the line and the facility is available to
make control setpoint changes on individual cells. When not in use the
terminals display any exceptional conditions that exist on the line allowing
operators to know at a glance which cells require attention.
In case of failure of the supervisory computer only this monitoring ability is

lost. The GELTROl units continue to function independently.
5. 13.4 The development/backup computer
This is a second mini-computer that is identical to the supervisory

computer and located in the computer room. It is used for program
development and as a backup in case of failure of the supervisory
computer.
Program development and testing is carried out on this computer so as

not to interrupt or interfere with the supervisory computer's activities.
When a program is ready for installation into the control system, it is
passed between the two computers through a standard communications
link.
The backup computer is also connected to the pot line communication

lines and data storage facilities. This way in case of failure of the
supervisory computer the backup can immediately take over all
supervisory functions.
5.14 Automatic Control Strategies

Although digital computers are able to collect large amounts of data very
quickly, the problems of lack of variety of information still remain. The
power of a computer system lies in the ability to develop a control strategy
based upon conditions that may be inferred from those variables that may
be measured and the history of the cells operation. The computer can be
programmed to take automatic action if appropriate or to alert operators
that manual assistance is required.
Whilst a computer can be used to automate many of the tasks, a different

strategy is required for each type of cell ( half break or point feed) and the
159
different operating conditions e.g. anode effect, high alumina
concentration and sick pot conditions. A specific program must be written
to implement each of these control strategies. The current strategy applied
for half break cells at Anglesey aluminium will be described in detail here.
5.15 Half Break Control Strategy
Anglesey's current half break automatic control strategy concentrates

upon maintaining a desired anode effect frequency and average cell
voltage. The temperature and condition of the cell are monitored daily by
potroom staff and the voltage and alumina set points adjusted based upon
the trends in these variables.
5.15.1 Automatic voltage control, AVe
The voltage drop across the aluminium cell is controlled by simple

feedback control using proportional action. A control block diagram is
shown in figure 5.2.
The voltage is sampled by the controller every 100 ms and the cell
resistance calculated from this and the line current. This value is known as
the Irawl resistance and is subject to high frequency variation. The raw
resistance is filtered through 1 and 3 minute digital filters to produce a
smooth resistance. Voltage corrections are made proportional to the error
of this value from a set point by movement of the anode beam. This beam
is moved by pneumatic motors that are operated by solenoids activated by
GELTROL. The voltage noise, leakage within the air pipes activating the
pneumatic motors and the inertia of the anodes reduces the accuracy of
the control such that small errors would cause unnecessary movement of
the anodes. To prevent this, a dead band around the resistance set point
is defined in which no anode movement is allowed to take place.
The operation of automatic voltage control is not always desirable,

however. There are different routines within GELTROl to control the cell
voltage under different operating conditions.
160
Cell voltag eV
Voltage + L Beam Anode beam
set po int Controller position
~-
movement
V
Figure 5.2 Voltage Control Block Diagram

5. 15.2 Normal operation
Under normal operation the smooth resistance is compared with the set
point every 3 minutes. If the absolute error is greater than the defined
dead band, typically 1.2 !J,Q, then the solenoid activation time is calculated
(1 )
where ts = activation time for the solenoid

Kp = constant of proportionality
£ = absolute error
td = dead time of the airlines, pneumatic motors and
solenoid.
The constant of proportionality is based upon an expected electrolyte
conductivity and the rate of movement of the anodes by the motor. As
movement is easier downwards than upwards, a different constant of
proportionality is used in each case.
In order to prevent abnormal conditions, such as anode effect or low

voltage, from causing the anodes to be removed from the electrolyte or
driven into the cathode, limits are placed upon the resistance that may be
controlled. These limits are expressed as a deviation from the target
resistance and are typically +1.6 !J,Q. Outside these limits, normal voltage
control, AVC, is not allowed and CElTROl may use an alternative routine
or call for manual assistance.
5.15.3 Noisy cells

If the frequency or amplitude of the voltage nOise exceeds predefined
values then the fluctuating error in the smooth voltage may cause
unnecessary movement of the anodes. This is overcome by increasing the
dead band in which no control may take place and by activating a noise
control routine.
It has been found that the stability of a cell is often increased by

increasing the ACD and this is achieved by increasing the voltage set
point. The noise control routine increases this set point temporarily and if
successful, gradually reduces the increase until the set point is back to its
original level.
161
5. 15.4 Anode effect
At the onset of anode effect, the cell voltage rises due to the increase in
polarisation potential and cannot be rectified by controlling ACO. Any
attempt at control by the normal AVC routine would cause the anodes to
be driven into the cathode. In order to prevent this, AVC is terminated if
the cell voltage exceeds an upper limit. If the voltage continues to rise and
exceeds 9V the cell is deemed to be in anode effect and an anode effect
termination routine is activated. This routine incorporates both alumina
feed and voltage control and is described in section 5.17.
5.15.5 Metal tap
The tapping of liquid metal from the cell lowers the pad depth so
increasing the ACO and hence cell voltage. The rate of voltage increase is
too great to be controlled by the normal AVC routine based upon a voltage
filtered over 3 minutes and use of this routine would likely result in the cell
voltage exceeding the upper limit of the controllable band.
The removal of metal from the cell is not directly detected by the controller
and any routine for controlling the voltage during tapping must be
activated by the operator. Use of normal AVC routines applied directly to
the unfiltered resistance at 18 second intervals has been utilised with
some success but often during the metal tapping operation the operator
controls the voltage manually.
Alumina is added to the electrolyte in batches of a fixed volume by

breaking the surface crust and dumping alumina into the bath from a
hopper suspended above the cell. The accuracy of this method is not
great with typical errors of +10-25 % by weight expected for a single batch
feed. It is not possible to accurately vary the amount of alumina fed and
control is therefore concentrated upon the alumina feed rate.
There is no means of direct on line measurement of alumina concentration

in the electrolyte and control aimed at maintaining concentration within a
specified band. As anode effect is the only measurable reference point for
162
alumina concentration, control centres upon the control of anode effect
frequency.
5. 16. 1 Normal operation
Under normal operation, the alumina is fed as a fixed batch weight at a

regular time. The amount of alumina added to the cell and the feed time
are designed such that the cell is purposely under fed so that anode
effects will occur at intervals and an excessive amount of sludge is not
allowed to build up at the cathode. The amount of alumina required for
electrolysis for a particular feed time is calculated by:
CE 1 I
- -.-. ·W ·t kg
100 2 z AI . F AI 2 0 3 f (2)
where Mf = alumina consumed by electrolysis kg

CE = current efficiency of the cell
I = cell current kA
tf = fixed feed time s
The cell may be under fed by either increasing the feed time or decreasing
the batch weight. Although this would theoretically give a regular and
predictable anode effect frequency, the error in the batch fed and
variations in current efficiency do not allow this in practice.
5.16.2 Search routine
The memory and rapid computational capabilities of digital computers

allows monitoring of the change of ohmic resistance with time of individual
cells. The trend of resistance with time and alumina content may therefore
be used to control feeding of alumina. The search routine relies on the fact
that, if all feeding on a cell is stopped, the cell voltage will rise as the
alumina concentration in the bath falls. As anode effect is approached the
voltage will rise sharply and this can be detected as a sharp rise in cell
resistance. If this rise exceeds a specified value over a defined sampling
period then the cell is deemed to be approaching anode effect and is fed.
163
Theoretically, this routine could be used to always detect the time at which
a cell should be fed. Unfortunately, abnormal conditions in the cell may
give rise to changes in cell voltage that can prematurely trigger feeding
and cause serious overfeeding problems. For this reason, the strategy is
subject to a number of restrictions that are designed to prevent this from
happening.
Demand feed strategy is designed to enhance the fixed feed strategy by

reducing the number of unscheduled anode effects in the under fed cell
whilst ensuring that overfeeding does not become a problem. The basic
feed time is retained and the Search routine activated only if the cell has
not had an anode effect within the previous 12 hours.
Whilst the controller is searching for rate of change of voltage due to the
decreasing alumina concentration there must be no movement of the
anodes to mask these changes. AVe is therefore temporarily suspended
whilst search is in progress.
Noise control and the metal tapping operation could also trigger feeding
and for this reason Search is temporarily suspended should these
conditions occur. As soon as the terminating condition is removed Search
is able to restart.
5.16.3 Starve routine
The starve routine is included in this strategy to ensure that the alumina
concentration and sludge content of the cell is unable to build up. This
routine forces a scheduled anode effect by suspending all alumina feeding
until anode effect has occurred and is only activated if the cell has not has
experienced anode effect within the previous 9 hours.
The combined effects of the demand feed strategy and starve routines is
to limit anode effect frequency whilst ensuring that the average alumina
concentration remains at a reasonably low level. Low levels of alumina in
centre fed prebaked cells, have been found by some researchers to be
associated with good current efficiency [1] and occasional anode effects
may be considered a better risk than the less controllable effects of too
much sludge at the cathode.
164
5.17 Anode Effect Termination
As anode effect is not terminated simply by the addition of alumina to the

cell the routine for anode effect termination consists of a sequence of
operations that are performed until the cell voltage has returned to a
normal operating level. This sequence uses aspects of both alumina
concentration and voltage control.
The anode effect termination, AET, routine is activated if the cell voltage
exceeds 9V. A typical AET routine is shown in figure 5.3. The routine
consists of alternately raising and lowering the anodes in order to disperse
the bubble layer, and the addition of alumina to increase the alumina
concentration. These operations may be repeated a number of times, but
if the failure location is reached then AET is deemed to have failed and
manual assistance is required. The success of the AET routine results in
an addition to the cell of double the standard feed weight of alumina.
5.18 Point Feeder Control Strategies
Point feeder cells are able to feed small quantities of alumina on a semi-
continuous basis and are therefore, in theory, able to maintain the alumina
concentration at a more constant level than a half break cell. The alumina
control strategy for a point feed cell is basically the same as that for a half
break cell with Search and Starve routines enhancing a fixed feed rate.
The short feed times, however, require that these routines are subject to
different rules to those in the half break routines.
The voltage control routine is the same as that used in a half break cell.
5. 18. 1 Search routine

The Search routine in a point feed cell is enabled if the cell has not had an
anode effect within the previous 9 hours and the time is within a
Search/Starve window. This window is designed to limit the search routine
to the night shift (midnight to 8 am) in order to minimise manual
disturbances to the feed routine.
165
0 1 2 3 4 5 6 7 8 9
0 BRAK LOWR 7.0 W006 JUMP 15 RAIS 5.0 DUMP BRAK
1 LOWR 5.0 W006 LOPO 1 6 -RAIS 3.0 WOOS RAIS 3.0
2 W006 BRAK WOOS DUMP DUMP DUMP DUMP
3
4 WOOS BRAK W010
5 DUMP DUMP DUMP DUMP

Sequence for Half Break Cell
0 1 2 3 4 5 6 7 8 9
0 DUMP BRAK LOWR 2.0 DUMP BRAK JUMP 13 LOWR 2.0
1 DUMP BRAK W001 LOPO 18 JUMP 24 RAIS 3.0 DUMP
2 BRAK LOWR 3.0 W001 LOP1 217 BRAK DUMP -RAIS 2.0
3 BRAK DUMP W002 RAIS 2.0 W002 RAIS 0.0 -W04S BRAK
4 DUMP DUMP
5

Sequence for Point Feed Cell
Nomenclature
BRAK activate crust breaker, both ends

DUMP activate alumina dump, both ends
LOWR n.n lower anode bus for n.n seconds
RAIS n.n raise anode bus for n.n seconds
JUMP nn jump to location nn if anode effect terminated
Wnnn pause nnn periods of 6 seconds = nnn x 6 seconds
LOPn xxyy loop to position yy allowed xx times if AE not terminated
eg LOPO 1 6 loops to 6. Loop is allowed only once
decision point
Figure 5.3 Anode Effect Termination Sequences

As point feed cells tend to operate at lower alumina concentrations than
half break cells the gradient of the alumina resistance curve at the
operating point is greater than for half break cells. The required rate of
change in resistance to terminate Search is therefore greater in the pOint
feed routine.
5. 18.2 Starve routine
The only modifications to the Starve routine for point feed cells are the
conditions required for activation. Starve will only occur if there has been
no anode effect within the past 48 hours and the time is within the
Search/Starve window.
5.19 Anode Effect Termination Routine
The basic form of Anode Effect Termination, AET, in a point feed cell is
the same as that in a half breaker. A typical sequence for a point feed cell
is shown in figure 5.3.
5.20 Alternative Strategies
The above strategy is that currently used at Anglesey and has been
developed over many years. Clearly there are alternatives, both in use
and under development. The alternative strategies reflect not just the
different types of cell but also different control philosophies. The basic
routines for feeding alumina and AVC are, however, much the same
throughout the industry.

There is some agreement[2,3,4,5,6,7] that the curve of current efficiency
against alumina concentration passes through a minimum and a typical
example is shown in figure 5.4. Current efficiency may be improved by
working at either high or low alumina concentrations.
In the former strategy, the alumina levels must be maintained whilst

avoiding excessive sludge formation at the cathode. In the latter,
166
95
Temperature 965°C
Weight Ratio 1.35
~-
- t \ 94
Calcium Fluoride
ACD
5%
4cm
~
CJ
c: 93
.!!!
CJ
w
==
1: 92
......::::J
Q)
0
91
90
1 2 3 4 5 6 7 8
Alumina Concentration (wr%)
Figure 5.4 Cell Current Efficiency vs. Alumina Concentration

concentration must be kept as low as possible without causing too many
anode effects. The particular strategy used depends upon the company's
philosophy.
The alumina feed strategies may be broadly split into two groups, open
loop and closed loop strategies.
5.22 Open loop Alumina Feed Strategies
Open loop strategies are the simplest and most easily implemented as the
feed of alumina is completely independent of the alumina concentration.
This is achieved by activating alumina feed with a simple timer device.
5.22. 1 Low alumina concentration
Open loop strategies may be operated at low alumina concentrations by

ensuring that the cell is always underfed. The problem with this approach
is that there is an estimated variance of +10-25 % in batch weight fed. If
the cell is operated close to the theoretical minimum the frequency of
anode effects becomes less controllable. In order to achieve a reasonable
anode effect frequency the average concentration frequently needs to be
much higher than the theoretical value.
The large batch weights of half break cells often give an average
concentration of around 4%, that is close to the minimum of the current
efficiency curve. The smaller additions of point feed cells reduces the
variation in alumina concentration but still cannot operate at the optimum
using this strategy.
5.22.2 High alumina concentration

High average concentrations may be achieved by overfeeding of the cell.
This strategy has been used in sidebreak cells where sludge settles along
the edge of the cathode and can help to stabilise the metal pad. A period
of overfeeding after an anode effect quickly increases the alumina
concentration. This is maintained by a period of slight underfeeding
followed by the forcing of an anode effect. This strategy is unsuitable in
167
centre break cells where the build up of sludge in the centre of the metal
pad has a detrimental effect upon cell operation.
5.23 Closed Loop Alumina Feed Strategies
Closed loop feed strategies attempt to control the alumina concentration

by inferring the concentration from cell resistance measurements. The
strategies described here are under development at Anglesey and aim to
operate the cells at low alumina concentrations.
5.23. 1 Half break cells: adaptive feed strategy
This strategy uses a continuously varying (adaptive) feed time controlled

by searching for anode effect before every feed. The search routine used
is the same as that in Anglesey's current strategy. To prevent overfeeding
of the cell by the termination of Search by conditions other than low
alumina concentration, feeding is only allowed within a window based
upon a base feed time.
The search routine is activated 40 minutes prior to the base feed time but
feeding may not occur earlier than 20 minutes prior to base. If 25 minutes
after the base feed time Search has not been successful then feeding
occurs anyway. AVC is suspended during Search and for 5 minutes
following a feed.
As a further safeguard against excessive overfeeding of the cell the

Starve routine is activated in this strategy if the cell has not experienced
anode effect within the previous 23 hours. This aims to produce an anode
effect frequency of one anode effect per day.
5.23.2 Point feed cells: adaptive feed strategy

A combination of two strategies are used to control alumina feed and
anode effects in point feed cells. These are an overfeed/underfeed
strategy (O/U strategy) and the use of search and starve routines. The
O/U strategy is designed to reduce the variance of alumina concentration
by alternately over and underfeeding the cell depending upon voltage
changes. The strategy has no means of determining the absolute value of
168
alumina and the search and starve routines are used to provide a
reference point.
The target range for alumina concentration in this strategy is shown in

figure 5.5. This is just off anode effect and the search routine is used to
find the lowest alumina concentration without the cell having an anode
effect. The O/U routine is then activated and takes over control of alumina
feed to the cell. Provided that there are no mechanical faults or long
interruptions to the control routines this strategy should eliminate anode
effects. In practice, however, anode effects are forced periodically by use
of the starve routine. This prevents overfeeding in the event of mechanical
problems or failures in the control strategy.
5.23.3 Search routine
The search routine is the same as that described in section 5.18.1.

Experience has found it advantageous to activate this routine every 9
hours. This helps to consume any extra feed the cell might receive such
as during an anode change.
A number of operations disturb the cell and following these, the search
routine is activated in order to find the appropriate alumina concentration
for the O/U strategy. These Searches are not subject to the Search/Starve
window restriction and will start 60 minutes after tapping or anode setting.
On successful termination of Search, the cell receives a feed cycle at both
ends. It will also trigger a change to overfeed and will suspend Ave for 30
minutes. This enables the voltage to stabilise and return to the level prior
to the end of Search.
5.23.4 Overfeed/Underfeed strategy (O/U strategy)

The O/U strategy uses a rise in cell resistance to indicate a fall in alumina
concentration i.e. underfeeding. Fall in cell resistance is correspondingly
used to indicate a rise in concentration i.e. overfeeding. Since this strategy
operates 24 hours a day it must be able to operate at the same time as
the AVe routine. This requires that voltage changes must be detected
before the AVe routine can move the bus to correct them. Voltage
changes are therefore detected within the AVe 'dead band ' and this band
169
-
!
~
c
~
-'iii -------+----
o I
I
I
I
I
I
I I
.-:
i \ Target Alumina
:
I
: Concentration
Anode
effect Alumina Concentration
Figure 5.5 Target Voltage for Adaptive Feed Strategy for Point Feed Cells
is used as the limits for under and overfeeding. A typical voltage trace for
the strategy is shown schematically in figure 5.6.
When cell voltage passes outside the dead band, AVe raises or lowers
the bus to bring the voltage back inside the dead band. These movements
are used to trigger over and underfeeding. Four consecutive raises
changes the feed from over to underfeeding. Two consecutive lowers
changes feed from under to overfeeding. In order to prevent frequent
changes on a noisy cell, no change is made if the four raises occur within
9 minutes of a lower or the two lowers occur within 9 minutes of a raise.
The cell is over or underfed by changing the feed time relative to a base
value. Typical setting for the amounts of over and underfeed are:
Underfeed 50% = (100 + 50) X Base feed time/1 00 (3)
Overfeed 30% = (100 - 30) X Base feed time/100 (4)
Additionally, any period of overfeeding is restricted to a minimum of 30

minutes and a maximum of 60 minutes, when the cell reverts to
underfeeding. After anode effect or changes between types of feed, AVe
is allowed but the OIU strategy ignores all raises and lowers for 12
minutes. Anode effects are often followed by several bus movements
before equilibrium is reached and these restrictions are designed to
prevent these extra movements from triggering underfeed too soon after
anode effect.
The standard AVe routine is designed to return the voltage to the centre
of the dead band. In this strategy, however, it is required to bring the
voltage back to only just inside these limits so that over and underfeeding
will control the voltage within the dead band. This is achieved by halving
the value of Kp in equation (1) used in the standard AVe routine thus
halving the bus movement called for by AVe.
5.23.5 Starve routine
The starve routine is used to prevent overfeeding due to mechanical

problems or failure of the control strategy. The routine is identical to that
described in section 5.18.2 and takes priority over search and
170
I
I
I
I
I
I
I
I
I
....... J ... . . . . .. . . . . . . . . . L..-_____ . . . . . . . . . . . . . . . . . .
I
I
I
I
I
I
CD
I
I
E
. . . . . . . .. ....;.._~ ... J ..... .
i=
I
I
I
I
I
"C
Q)
~
Q)
"C
c:
=>
overfeed/underfeed routines. This routine is activated if there has been no
anode effect within the previous 48 hours. When the feed routines are
working well, all anode effects will be due to Starve giving an average of
0.5 anode effects per pot day. Test cells have been run successfully with
one anode effect every 10 days but this requires consistent alumina feed
and good bath level control.
5.24 ACD Control Strategies
ACO control strategies aim at achieving an optimum energy efficiency. For

any cell operating condition there is a critical ACO below which there is
increased loss of current efficiency and above which there is increased
energy consumption.
5.24.1 Fixed ACD
Constant ACO routines assume that the bath composition and

temperature is essentially constant. Under these conditions, the optimum
ACO is fixed. By further assuming that the rate of rise of metal pad level is
equal to the rate of electrolysis of carbon anodes, once the optimum ACO
has been attained there is no need for further anode movement. This
strategy is therefore only suitable for point feed cells where voltage
fluctuations due to alumina feeding are reduced.
Ouring the tapping operation and after anode effect terminations, the ACO
is returned to its optimum value by returning the cell to a target voltage.
5.24.2 ACD optimisation

This scheme utilises the relationship between anode-cathode distance
and electrical noise generated by the cell. Electrical noise increases with
decreasing ACO and becomes excessive when the point of voltage
instability is reached. At this point current efficiency begins to be seriously
decreased.
This strategy assumes that this is the point of optimum ACO and controls
voltage just above the point of instability. Periodically, the anodes are
lowered until noise becomes excessive and the voltage setpoint adjusted.
171
Voltage control is then by standard AVC routines. This method of control
is not currently used at Anglesey Aluminium.
5.25 Summary
This chapter has discussed some control strategies than are used in
industry and that can be investigated in simulation. The next chapter will
discuss the programming of the aluminium cell model and the control
systems and will compare the results of the simulation to experimental
data in order to validate the model.
References

2 Gjerstad et ai, Symposium on Industrial Electrode Processes with Gas

3 Belyaev et ai, Izv, Vyssh. Ucheb. Zav., Tsvetn. Met., vol 4, no 3, 1961,
p67
4 Grjotheim K. et ai, Can. Met. Quart., vol 11, 1972, p295
5 Firsanova et ai, Izv. Vyssh. Ucheb. Zav., Tsvetn. Met., vol 5, no 3, 1962,
p52
6 Szeker, Acta Tech. Acad. Sci. Hung., vol 10, 1955, p91
7 Bersimenko et ai, Tsvet. Met., vol 46, no 3, 1973, p30
172
6.0 Dynamic Simulation of an Aluminium Reduction Cell
One of the main requirements of the simulation is that it should be able to
be used both on-site at Anglesey Aluminium and at Newcastle University.
With the advent of cheap powerful microcomputers this presents no
problem and most of this work was carried out on a Compaq 32 bit IBM
compatible personal computer. A wide range of programming languages
are available for these machines and the choice of language is based
upon the applications for which the simUlation is intended.
6.1 Choice of Programming Language
The simulation is required to predict the dynamic behaviour of the cell over
long periods of time. In order that this takes as short a run time as
possible a compiled, rather than interpretative, language must be used.
The simulation of different control systems and types of cell requires that
parts of the model be altered periodically. It is therefore desirable that the
language used allows the program to be specified as a number of
modules eg process model, control system, user interface etc. If changes
are made to any module, only that module need be re-compiled not the
whole program.
The program was originally written in a version of PASCAL that allowed a

degree of modularisation. This was successful in simulation but attempts
to produce a real time version with communication to the Anglesey
Aluminium control computers, CElTROl through an industrial signal
processor encountered problems due to the relative speeds of the
simUlation and communication.
A solution to this problem was the use of a language with multi-tasking

capabilities that allows different routines in the programs to be given
different priorities. Modula-2 was chosen as it is a language that has
evolved from PASCAL and transition between the two was relatively
simple. Further details of multi-tasking and real time simulation are given
in chapter 7.
173
As well as providing multi tasking capabilities, Modula-2 has a number of
other advantages over PASCAL. These are as follows:
1. Increased Modularisation: Modula-2 programs consist of a number of

user defined and library modules that are linked together after the
compilation stage. This has been utilised here to build up libraries of
control and model routines. Different control strategies may be easily
simulated by choosing the relevant routines from the library.
2. Model Security: Interfacing between individual program modules is by

Definition modules. These contain information of the objects (eg
Procedures, Variables) available from the program module. This allows
objects to be used by a model builder without the necessity of source
code being available.
3. Debugging: The use of small modules in developing large programs

greatly reduces debugging and significantly improves software
management. Each module can be individually tested and debugged and
any faults in the program more easily tracked to a specific area.
4. Portability: The modular structure of the language greatly enhances

the portability of programs between different machines. There are no
predefined liD statements, maths functions or string manipulation routines
in the Modula-2 core. All such routines are imported from library files that
may be produced specific to a particular machine.
6.2 Initialisation
The complexity of the overall process model does not allow for all of the
variables to be initialised by steady state analysis. Initial conditions were
therefore obtained by running the simulation with the approximate starting
conditions and allowing the program to converge to a steady state. The
process variables at this point were obtained as a memory dump from the
program and provide the 'steady state conditions of a 'standard ' run. This
I
is then used as the starting condition for subsequent simulations.
Once this start point has been achieved, memory dumps of the process
variables represent the system at any point in time. The conditions at the
174
end of one simulation can therefore be used as the starting point of
another simulation. In effect, the simulation can be continued without
interruption.
This feature can be utilised to investigate the effects of changes in

operational parameters relative to a standard run. An example of this is
the simulation of the effects of changing to a different type of alumina. The
memory dump of a standard run is used as the start up conditions but with
the parameters pertaining to alumina type altered ie thermal conductivity,
dissolution rate constants, specific heat capacity etc.
Separate memory dumps are produced for the control and model
variables. In addition, the control variables may be initialised from
complete or partial memory dumps. The complete version contains all
historical data for the control system and allows the simulation to be
continued from the point at which it was stopped. Use of the partial dump,
however, allows for changes to be made in control parameters and
simulates the resetting of aGELTROl device.
In addition to initialisation of the simulation from file, many of the operating

conditions may be changed from a menu system in the program. This
makes the software relatively easy to use by operators who have not been
involved in the programming and is an important aspect of the program
design if it is to be of use within the aluminium industry. Full details of the
menu system developed are given in Appendix A.
6.3 Presentation of Results

The simulated conditions within the cell are continuously updated on
screen as the simulation progresses. The screen format is designed to
display those operating and control parameters that are normally available
for a working cell. This allows the user to monitor the progress of the
simulation and recognise situations that can occur in the working cells.
The simulation allows for some operator settings, for example target
voltage and metal tapping, to be changed interactively as conditions within
the cell change. The output screen for the program is shown in appendix
B.
175
The simulation results are also stored in ASCII format files on the
computer disk. This produces vast quantities of data that must be
processed. The best means of interpreting the data is graphically but the
commercial spreadsheet programs available at the time of this work were
unable to deal successfully with the amount of data produced. It was
therefore necessary to produce a graphics package to do this. The
program allows the user to produce graphs of combinations of parameters
and includes facilities to zoom inion areas of interests. Hard copy is
I
provided on a Hewlett Packard plotter and full details of the package are
given in appendix C.
6.4 Numerical Solution of Differential Equations
This simulation uses Euler's method for the solution of differential

equations. This is a first order method and as such requires a much
smaller integration time to achieve the same accuracy as a fourth order
predictor method such as Runge Kutta.
Although the program is required to simulate cell operation over long

periods of time the main CElTROl routines work on a one second interval
and anode movement is calculated to one tenth of a second. In addition,
anode effect termination sequences can effectively limit the duration of
anode effect to one minute. An integration interval of around one second
is therefore required to successfully simulate all aspects of cell operation.
With such a small integration interval, the accuracy afforded by Euler's

method is perfectly adequate and has the advantage of being much
simpler and quicker than a higher order method.
6.5 Simulation of Current Control System (CElTROl)

The CElTROl control algorithms are programmed in a derivative of
FORTRAN and downloaded into the read-only memory (ROM) of the
CElTROl computers. To simulate the operation of the aluminium cells the
CElTROl algorithms must be converted into Modula-2 and linked into the
physico-chemical model of the cell. As the behaviour of the cell cannot be
separated from the effects of the operating system an accurate simulation
176
of the control algorithms must be produced to allow comparison of the
simulation with measurements from the cell.
Initially the full control program for the operation of half break cells ,
including the starve and search routines, were programmed as a Modula-2
module exactly as they were implemented in Modcomp FORTRAN. This
implementation of GELTROl was used as the 'standard ' version for
verifying and tuning the simulation.
6.6 Validation of the Simulation
A large number of unknown variables such as average heat transfer

coefficients are unique for individual cells and are very difficult to
determine. In order to validate the model, a 'standard ' run will be defined
with variables tuned to give the average conditions expected in operation
of aluminium cells.
The complete set of initialisation variables used in the standard run are
listed in appendix A and the results of a 170 hour simulation shown in
figures 6.1 to 6.6. The simulation is for a mark IV half break cell operating
at a target voltage of 4.39V with Anglesey Aluminium's search and starve
routines activated. Alumina is fed at a rate of 90 kg on a base feed time of
67 minutes. The current through the cell is constant at 151 kA and the cell
is configured for eighteen 53" anodes. A comparison of the average cell
performance for the first half of 1986 with the simulated performance is
shown in table 6.1.
The voltage trace for the simulation is shown in figures 6.1 and 6.2. This
shows an anode effect frequency of 1.55 AEs per day with anode effects
occurring at 7, 16, 35, 53.5, 64.5, 83, 101.5, 121, 139.5, 158 and 169
hours. Those occurring at 7, 35, 53.5, 83, 101.5, 121, 139.5 and 158
hours are due to the starve routine and successful searches have
occurred at 28.3, 47.5, 77.2, 95.6 and 151.5 hours. The average voltage
over the simulated period is 4.41 V the deviation from target voltage being
due to anode effects. This voltage is that measured by GELTROl and is
not directly comparable with the gross operating voltage which includes
the voltage drop across the buswork between individual cells. This
accounts for approximately 0.03V at 151 kA and gives the simulated gross
177
Parameter Actual Simulation
Energy Efficiency (DC kWh/kg) 14.8 14.6
G ross Voltage (V) 4.48 4.44
Current Efficiency (%) 90.1 90.5
Anode-Cathode Distance (cm) 4.85 4.68
Bath Temperature (OC) 969 971
Wt Ratio NaF/AIF3 1.22 1.23
Table 6. 1 Comparison of Simulated Performance and Actual

Performance for a Half Break Cell in 1986
6.00
I
I
(j)
tJ
~~
r-
o
> ~
....jl ""
I
''''111 " -J
~
~~
r'u.. L
-.....vvI " ~~
... "
~
_
~
u.
-- ~ ~~ ~~u~
3.00 , , , , I j
o 17 34 51 68 85 102 119 136 153 170

Time (hrs)
Figure 6. 1 Simulated Voltage Profile for 151 kA Centre Break Cell

6 . 00)
_ .- Target
...
o
> .... . .A .l l
3 .0 )
o 4 8 12 16 20 24 28 32 36 40 44 48
Time ( hrs )
6 . 00)
~
u
o
> ./. A I .A L
3 .00
50 54 58 62 66 70 74 78 82 86 90 94 98
Time (hrs)
6 . 00)
.
u
o
> ~ . . I ! . -"
3.0 0
100 10 . 1 09 11 2 116 120 124
T1me ( hrs)
128 132 136 140 , .. 149
Figure 6.2 Simulated Voltage Profile for 151 kA Centre Break Cell
operating potential as 4.44V. This compares very favourably with the
average value for the first half of 1986 of 4.48V[11.
6.6. 1 Current efficiency
Metal tapping was set to a 24 hour tap rate with a target pad of 14 cm.
The first tap occurred at 22 hours and the total mass of metal tapped over
the 170 hour period was 7980 kg. The amount of metal is used by the
industry to calculate the overall current efficiency over a period of time.
The calculation is based upon the assumption that the metal tapped is
equal to the net amount of metal produced over the period giving
CE I
Mtap - -. ·W·t (1 )
100 Z.F
AI
AI
where Mtap = total metal tapped over period kg

I = average cell current over period kA
t = time of period s
rearranging and inserting values for W AI, ZAI and F gives the overall
current efficiency
M
CE = 1.072x106.~ (2)
I· t
The current efficiency by this method for the seven day period simulated in
the 'standard' run is then 93.7 % •
This value for current efficiency is somewhat higher than that for
commercial cells at Anglesey Aluminium which operate at an overall
efficiency of about 90%. The discrepancy between the two values is due
to the short time period of the simulation. In the industry, the current
efficiency is calculated as quarterly and yearly averages. This removes
discrepancies caused by the method of tapping.
Although the aim of tapping is to return the pad depth to a target level the
amount tapped is not directly proportional to the deviation from this set
point. The schedule used for determining the metal tapped is shown in
table 6.2
178
Deviation from target Metal tapped
em kg
deviation at last tap < 0 0

>=0 900
0 1140
1 1320
>1 1500
Table 6.2 Aluminium Tapping Schedule at Anglesey Aluminium

The deviation from target pad is measured using a metal rod to a
maximum accuracy of 0.1 cm. The value is then rounded to the nearest
integer. Unless the metal pad at the time of tap is always at the target
value, it follows that the pot will have periods when it is effectively over or
under tapped. The frequency and duration of these periods will depend
upon the actual current efficiency of the cell. The use of this method of
calculation can therefore produce misleading results if the period in
question is short.
An additional source of error is the use of depth, rather than volume, to

determine the desired tap weight. Fluctuations in the thickness of the
frozen sidewalls results in the depth corresponding to different volumes
and can cause errors in the amount tapped. Over a long period of time
these factors become negligible and the calculated current efficiency
approaches the average current efficiency of the cell.
An alternative method of calculating the overall current efficiency of a cell

is by taking the average of the instantaneous current efficiency over the
period in question. This may be measured experimentally by gas analysis
using the Pearson Waddington equation[2] but this method is not suitable
for wide scale use on industrial cells. Additionally, the presence of gases
other than carbon monoxide and carbon dioxide in the sample introduces
errors in the calculation and for this reason it is not suitable during anode
effects. The simulation, however, calculates the instantaneous current
efficiency based upon the back reaction of aluminium in the melt and the
results for the 'standard ' run are shown graphically in figure 6.3. The
average value for the 170 hour period is 90.4 % which compares
favourably with the expected 90 %.
6.6.2 Energy efficiency

The energy efficiency of the cell is defined as the energy required to
produce a kilogram of aluminium and is normally quoted in DC kWh per kg
of aluminium. This is determined from the gross cell voltage and current
efficiency by the relationship
Energy Efficiency = 298 1x Gross Voltage (V) DC kWh kg- 1 (3)

. Current Efficiency (%)
179
..
w
u
50 LI____ ~ ____ ~ ______ ~ ____ ~L_ ____ ~ ____ ~~ __ ~ __ _L~ ______ ~ _ __L_ _ _ _ _ _ ~ ____ ~ ______ ~ ____ ~L_ ________ ~ __ ~ __ ~ ________ ~ ______ ~
o 17 34 51 68 85 102 119 136 153 170

T 1 me (hrs)
r-l __ -
6.00I
~~ ~r--~
-- --..- ...-- l 1 r---'-
E
u
o
u
<
0 .0 J
o 17 34 51 68 85 102 119 136 153 170
T1me (hrs)
8 . 00
...o
N
<
. .
:t
o oo LI______~______~______~____~______~____________________~______~____________________~______~--------------------~------~--------------------~
o 17 34 51 68 85 102 119 136 153 170
T1me (hrs)
Figure 6.3 Simulated Current Efficiency, Concentration and ACO Profiles for 151 kA Centre Break Cell
The value for the simulation is therefore 14.6 DC kWh I kg which is close
to the actual value for the mark IV cell of 14.8 DC kWh I kg.
6.6.3 Cell temperature
The temperature of the electrolyte is shown in figure 6.4. Also shown is

the liquidus temperature of the melt. The overall trend of the temperature
is constant with fluctuations about an average of 971.4 °C. The
temperature attained at anode effect is approximately 990°C and varies
with both duration and maximum voltage of the anode effect. The
temperature follows a consistent pattern during a feed cycle with a sharp
drop at the beginning of the cycle corresponding to the addition of cold
alumina followed by a rise in temperature to the end of the cycle. The
average temperature over the cycle increases as the cell approaches
anode effect.
6.6.4 Weight Ratio NaFIAIF3
The weight ratio of NaF/AIF3 is shown in figure 6.5. The graph shows a
sharp rise in ratio at anode effect corresponding to the evolution of
gaseous fluorides by the anode effect reaction and also an increase in the
rate of volatilisation due to increased temperature. The addition of bagged
aluminium fluoride to the cell is characterised by a sharp fall in ratio. Over
the period of simulation, the ratio is stable and fluctuates about an
average of 1.23 0/0.
The average values for temperature, 971°C, and Ratio, 1.23, obtained by
simulation both compare favourably with the historical data for the year
1986, 969°C and 1.22 Ratio, although the simulated temperature is 2 °C
greater than the actual value. This is due to the slightly higher value of
Ratio which raises the liquidus temperature hence requiring a higher
temperature to maintain the heat lost through the sidewalls of the cell.
6.6.5 Anode to Cathode Distance (ACD)

The values for anode-cathode distance show some apparent discrepancy
with a value of 4.68 cm obtained by simulation compared to 4.85 cm for
the actual value. Anode-Cathode distance is not measured directly on
180
0
. ~- . --
=-=--
-.: . ~ r--..
r-I
r0
L.[)
r-I
\0
r0
r-I
...:::J
Q)
-...ctI
Q)
(j)
r-I
a.
E r-I
~
en
:::J N
"0
:::J 0
0- r-I
:.:J
L.() v
ro (1)
E
'r
ro I-
\0
o
o o
o avln+t:?vladwaJ..
o
Ii
1. 25
o
..
.
at:
:E
1.20 ~1 ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~
o 17 34 51 68 85 10 2 119 136 153 170

T i me ( h rs)
1000
u
..
L
....
::>
..
L
.....
Co
E
950 LI____ ~ ____ ~ ______ ~ ____L __ _ _ _ ~_ _ _ _~_ _ _ _~_ _ _ _~_ _ _ _~_ _ _ _ _ _~_ _ _ _L __ __ _~_ _ _ _~_ _ _ _ _ L _ _ _ _~_ _ _ _~_ _ _ _~~_ _ _ _~_ _ _ _L __ _ __J
o 17 34 51 68 85 102 119 136 153 1 70

T1me (hrs)
Figure 6.5 Simulated Ratio vsTemperature for 151 kA Centre Break Cell
commercial cells but is calculated from the voltage drop across the
electrolyte by the relationship
(4)
where Ve = voltage drop across the electrolyte

k = electrical conductivity of the electrolyte
f = anode cathode distance
I = current flow through cell
Ab = effective area through which current flows
The relationship between the gross operating voltage and the voltage drop
across the electrolyte is given by the equation
(5)
where Vcell = gross operating voltage

E" = equilibrium potential
11 = total overpotential at the anode and cathode
V ext = voltage drop external to electrolyte
The calculations for the actual and simulated values differ only in the value
used for the overpotential. In Anglesey Aluminium's calculation, the
overpotential is calculated from a Tafel expression for the reaction
[1 ]
based solely upon the current density at the anodes. The effect of alumina
concentration is disregarded and the relationship appears to have been
determined for a concentration of 5 wt% alumina.
The simulation, however, includes concentration and temperature

dependence of overpotential with the effect that overpotential increases
with decreasing alumina concentration. The proportion of gross voltage
attributable to the ohmic drop across the electrolyte is therefore decreased
resulting in a smaller anode-cathode distance. Additionally, the value
determined from simulation is the average over the period and includes
181
30.00 rl------------------------------------------------------------------------------------------------------------------------------------,
E
u
VI
...a.
~
.
c
Bath
10.00 1L-_____ L_ _ _ _ _ _ ~ _ _ _ _ _ L _ _ _ _ _ _~ _ _ _ _ _ L _ _ _ _ _ _~ _ _ _ _ _ L _ _ _ _ _ _ ~ _ _ _ _ _ L _ _ _ _ _ _~ _ _ _ _ _ L _ _ _ _ _ _ ~ _ _ _ _ _ L _ _ _ _ _ _ ~ _ _ _ _ _ L _ _ _ _ _ _ ~ _ _ _ _ _ L _ _ _ _ _ _ ~ _ _ _ _ _ L _ _ _ _ ~
o 17 34 51 68 85 102 119 136 153 170

Time (hrs)
10.00
.
u
..,'"
-'
.
..,
.
"-
5 OO LI____ ~ ____--'______ ~ ____ ~ _____'_____ ~ ____ ~ ____ ~ ______'______ ~ ____ ~ ____ ~ _____'_____ ~ ____ ~ __ --~------'------~----~-----'
o 17 34 51 68 85 102 119 136 153 170

T1me (hrs)
Figure 6.6 Simulated Bath and Pad Depths for 151 kA Centre Break Cell
the very low anode-cathode distances due to anode movement during
anode effect termination.
6.6.6 Parameter trends
The 'standard' run shows good agreement between the simulation and
industrial data and can therefore be used as a reference for the validation
of the more detailed dynamic behaviour of aluminium reduction cells. The
lack of comprehensive data for individual cell requires that the model be
tuned to average conditions at the Anglesey plant. In validating the effect
of disturbances to operating conditions it is the resulting trend in the cell
parameters that is of importance rather than the absolute values attained.
Simulation results will be compared with both experimental data and the
results obtained by other models.
6.6.7 Alumina concentration
The dynamic variation of the concentration of alumina dissolved in the

electrolyte shows good agreement with both industrial data and the results
of other simulations. A graph of the first 24 hours of the standard
simulation is shown in figure 6.7. Actual concentration profiles have been
measured using an alumina concentration meter and chemical analysis[3]
for a 100 kA Soderberg cell and the results are shown in figures 6.8 and
6.9. This type of cell is not directly comparable with the pre-bake centre
break cells simulated here but the shape of the curve representing
dissolution of alumina and its consumption by electrolysis is similar.
The results of measurements for a 150 kA centre break cell with a two
hour feed cycle[4] are shown in figure 6.10 and concentration curves from
the simulation by Ek and Fladmark[5] in figure 6.11. These all have similar
shapes over individual feed cycles.
Another point of comparison is the alumina concentration at anode effect.

The concentration at which anode effects occur in the simulation is
1.5 wt% and this is the average concentration found using the alumina
concentration meter. Anode effects in figure 6.7 also occur between 1 and
2 % alumina. However, the concentrations measured in figure 6.8 are
much lower at about 0.5 % whilst those simulated by Ek in figure 6.11 are
higher at about 2.5 %.
182
6.00
M
o
N
r-
«
~
+J
3
l .ool~~~~~~~~~~~~~-L-L-L~~~~_ _L-L-~~~~~-L-L~~~
3.2 6.4 9.6 l2.8 l6.0 19.2 22.4

0.0
Time (hrs)
Figure 6.7 Alumina Concentration over First 24 Hours of the 'Standard'Run

o
N
'"'-:J
o
.r:
o •
- <I)
E
<J CO
<J
~6
.-'"
'">- '-
ro <I)
c '-'
<! <I)
<J: -ro
"l:
ro
$Y~O
u
E
.-Ec
<I) :J
.r:
u
-<!
0 <J N
o
o
N
~ Co
III
c: ...,CIJ
<l: CIJ
::E
~
III III
U
.-Ec:
E
CIJ :J
-
CO
. "----CL OJ ..c
u
o <J
~
<l:
<J .
-
-.D
~eaJ8 ~ ~6J-'
,-<P <J
'.~~-~~~
o
N
Dutil Temperature, o C Current Efficiency, %
lD lD lD lD (0 lD
~ U1 U1 (j) (j) -...J 00 00 00 to
U1 0 U1 o U1 () -...J 'Y' to 0
'\
00
.~ r>
?
o
o \
:-rJ
.~
cO"
c:: '"'
(t) ;:)
~
)
o
!J)
-.&.
o \
a I-"
N \•
JJ o .~
CD
g.
C"')
......
::>
1:1
:::: ,>nn \ Ji. [>
Effect
0" N /." i ::=--" <-
:::l o
o
.\
~ •
.
[>U---::-O
~
:::::::
A
o · '; \
/~ s:E:
CD
ti3
o
'" f>l> _cr
~ ~:'> ~~O/"
......
0" O'l Anode Effect
:::l
~"
o
c ( J>--U
S '\ . y-a-CYcY
o~
00
o
I
o .'- :;;= t>~ /\node F.f fect
~
c::
....,
OJ
I-"
o
o
o
I• "
\
• y~-1:::U
(t)
III
I-"
Iv \
~ \
;:)
;:)
•
'c> ~
~
>
CD
CD :.:
\.
Q..
~
N
o
I' '. "--
\.~
/L>~·D.-'
C"')
<t~
o
r'~ D»I> o.~.U Anode Effect
A
o
o
.(\ ~.
~ ? 0"
.<"0 Metnl
. ....~ L.J
(j)
;:)
Cell Voltage
'\
.~. ~ [> J ~.
~
o ~ ... \.Y' AI umi na,%
~
m
I~. 5 r ------,----,---.------__
4.4
4.00n-----------7G---------~-----------~
12 18
t (hr)
The cell voltar.e.
10.0r----------r--------~-----------
7.0
6.0~-----J---- _ _ _ L _ _ _ _ _ _~
o 12 18
t (hr)
Thic~~ess o~ unpcr freeze.
5.0r-----------~r-----------~--------------
p (::)
ox
3.0
2.0~ __________~~----------~------------~
o 6 12 lR
. t (hr)
Consentration 0: Al 0 in
2 3
t~e bath
Figure 6. 11 Simulation of Alumina Concentration, Freeze Thickness and

Cell Voltagel5J
As anode effect occurs due to the current density at the anode exceeding
the limiting current density for the reaction, the concentration at which
anode effects occur will vary. This is reflected in the occasional triggering
of an anode effect by the changing of an anode. Removal of an anode
from the cell reduces the area for current flow and increases the anode
current density. If the alumina concentration is low this may cause anode
effect. The disparity between the absolute values of anode effect alumina
concentrations may be due to operation of the cells at different current
densities. This information has, however, not been supplied by the above
authors.
The effect of an anode change on the alumina concentration in a 150 kA

centre-break cell is shown in figure 6.10 at around 9:00 am. This shows
the addition of alumina causing a rise of about 0.7 % in alumina
concentration. The increase in alumina content of the bath is dependent
upon the anode change operation and includes a highly random element
as the crust and sidewalls are broken into the cell during the removal of
the spent anode. Anglesey Aluminium estimate an alumina addition during
anode change of the equivalent of about one dump of ore. The simulated
effect of anode change is shown in figure 6.12 and also shows a rise in
alumina concentration. The size of the simulated rise is only one tenth of
that measured and this reflects the small mass of alumina specified as
being added to the bath due to anode change.
6.6.8 Cell voltage

The dynamic behaviour of cell voltage also compares very well with that
obtained from voltage traces of half break cells operating under Anglesey
Aluminium's control strategy. During a feed cycle the voltage is controlled
within the dead band until automatic voltage control (AVe) is switched off
prior to a feed. The voltage then starts to increase with the rate of change
of voltage dependent upon the concentration of alumina in the electrolyte.
In figure 6.13 the voltage change through an anode effect is compared

with the results of a 3 minute voltage trace of cell 2A 18. The cell was
operating at a current of 152.3 kA and a base feed time of 68 minutes.
The characters in the trace denote the following voltages:
183
5 . 00 1
Anode
Change
1 . 00LI______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______L __ _ _ _ ~_ _ _ _ _ _~_ _ _ _~_ _ _ _ _ _L __ _ _ _~
85 . 0 85 . 2 85 . 4 8 5 .6 85.8 86.0 86 .2 86 . 4 86 . 6 86.8 8 7. 0

Time (hr s )
20 . 00
.s::
\
.,
c-
..,
Ol
..
.s::
.,
II)
w
Anode
Change
E ~
15 0 0
85 0 85 .2 85 . 4 85 . 6 85 . 8 86 . 0 86 . 2 86 . 4 86 . 6 86 . 8 8 7. 0
Time (hr,)
Figure 6. 12 Simulated Effect of Anode Change upon Alumina Concentration

Day 0 0: 0
4.02 4 .12 4.22 4.32 4.42 4.52 4.62 TMR FED EVENTS
0: 0 : . . . . : . . . . : . . . . : . . . . : . . . . : .. .. : . . . . X. Z .. : . . . . : . . . . : . . . . :
'" ........... 25 68
0: 3 x Z 28 68
0: 6 TX SR: 31 68
0: 9 T XZ 34 68
0:12 T Z 37 68
0:15 T ZR 40 68
0:18 T ZR 43 68
0:21 T Z 46 68 L
0:24 T X S R 49 68
0:27 T SXR 52 68 L
o : 30 : . ... : .... : .. . . : . ... : .... : .... : .... T .. . . : .. . . : SRX. : .. .. : .... : .... : .... 55 68
0:33 T S R X 2 68 A
0: 36 X S R T 5 68 A
0: 39 X T R S 3 68
0:42 X Z T 6 68
0:45 X SR T 9 68 R
0:48 X S R T 12 68 R
0:51 SR T X 15 68 R
0:54 Z T X 18 68
0:57 Z: T X 21 68
1: 0 : . . . . : . ... : . . . . : . . . . : .. .. : ... SR . .. . T. X .. : . . . . : . . . . : . . . . : . . . . : . . . . : . . . . 24 68
1: 3 Z: T X 27 68
1: 6 Z: T X 30 68
1: 9 SR T X 33 68
1:12 SR T X 36 68
NOTATION
TMR Feed Timer FED Feed Time Dead Band
T Target Voltage X Simulated Voltage
S Actual Smooth Voltage R Actual Raw Voltage Z Coincident Point (S R Xl
EVENTS L Lower Anodes R Raise Anodes A Anode Effect
Figure 6. 13 Comparison between Simulated and Actual Voltage Trace During

Anode Effect in Cell 2A 18
Xsimulated voltage
Ttarget voltage 4.32 V
Sactual smooth voltage ( 18 s filter)
Ractual raw voltage ( unfiltered)
Z common value for two or more voltages
: voltage dead band
There is good agreement between the rate of change of voltage prior to

anode effect as well as in the duration of anode effect and the time taken
for the voltage to return to target.
During anode effect termination, there are a number of rapid changes in

cell voltage due to anode movement and noise. The large sampling time
(3 minutes) of the trace can miss this effect and since the simulated and
actual traces are not fully synchronised, this accounts for the larger
discrepancy during the 36-45 minute period.
The voltage change during anode effect is shown in more detail in figure
6.14. The experimental data used here is from a 10 second scan for cell
1A3. The simulation again shows good agreement with the industrial data.
As is expected, the tapping operation causes an increase in cell voltage

due to the lowering of the metal pad. In the simulation this is controlled by
a CELTROL routine and results in a voltage peak of about 5 V. On site,
this routine has been disabled and the voltage during tapping is controlled
by the operator. The variation in voltage during the tapping operation can
therefore increase or decrease as is shown in the resistance trace in
figure 6.15.
6.6.9 Bath depth
The level of molten electrolyte in the cell is primarily affected by the

temperature of the cell and by movement of the anodes. An increase in
temperature in industrial cells generally results in an increase in bath
depth due to melting of frozen electrolyte from the surface crust. The
simulation predicts this trend as is shown in figures 6.16 and 6.17. The
temperature in each of these runs is changed by introducing a step
change in the target voltage. Figure 6.16 shows the continuation of run
AI6c3 with a step change from 4.39 V to 4.29 V and figure 6.17 the effects
of a further step change from 4.29 V to 4.49 V.
184
10
- - - Simulation
9 - - - - - - Target \()Itage
8 o measured
-->
Q)
7
C)
as
==0 6
>
5
JIIIIIIIIIIIIIIIIIIIIIIIII~_- - - _L - -- - -- - -- - -- - -- - -- - -- - -- - -- - -- - -- - -- - --
3
29 31 33 35 37 39 41 43 45
25 27
Time (minutes)
Figure 6. 14 Voltage Trace of Anode Effect Predicted by Model

21. 0
L .
20. 4
. .. .
19.8
. . 19.2
• I •
RES I ST MICE MICRO OHMS
1 8. 6 18. 0 17. 4
• ,
16. 8
. L
16,
•
~ 15. 0
.
~
: tt '{I
.J
CO
\ ('tJ
q::
(()
(D
"
"a"-
N
N
' 41
...,t-..
~
'-
...
: t€ '{I
-
.,--
0
0..
'+....
r 0
I- .... 0')
a .:I ~
a ·S
(T) '", Q
.;;
'4
'4. ... : ~l '{I Q
~
., (
~
N
-
I\J
~
'(
I)
~ 0')
,\ .c::
\., ... t::
• f'- ., : .1 '{I .g
Q)
(D ()
~
(T)
<
~
: . '{I Q)
I- ()
a ~
-c::
'-
ll. It: r
~ ~ ctI
oc
a ~~
,. ... .~
ll.1 CJ)
'tI
-"
Q)
.J i~
:r '
(
16 ---'r "K'i!1
ex:
" a... 1- l{)
.,--
J
•, uI I
•o ·
0~ '
eO
-
\0, n::
Ul
(J)
: tt ~I
@
::J
.0')
t
LL
.;:
..., t:
(J)
...
w : t€ 'i!l
a:: .
"'«
<.
a::
I
I • r tl '(I
0'12 v'02 8 ·61 2 "61 9 "8t 0"81 po ·L 1 8"91 2"9t 9 "S l o 'S t
S~HO O~Jlt4 3JNY1SIS3~
E
u
.
r:
0-
o•
.
r:
.·
15.00 ' L__________~~____~__________~____~____~____~~----~----~----~----~~--~----~:_----~--~;;--__~____~~--~----_::_----~----~»~----~--__;

o 4 8 12 16 20 24 28 32 36 40 44 48
Time (hrs)
r --------------------------------------------------------------------------------------------------------------------------------------------------------------------------,
10.00 '
•
"•"
.J
·
".
S.OO L'____________ ~ ______ ~ __________________ ~~ ____ ~ ______ ~ ____ ~ ____________ ~ ______ ~ ____ ~ ______ ~ ________________________________ ~~ ____ ~ ______ ~ ____ ~ _____________'
o 4 8 12 16 20 24 28 32 36 40 44 48
Time (hrs)
98S 'r----------------------------------------------------------------,---------------------------------------y---------------------------------------------------------------,
·
L
~
·•"
L
·
(
S' IL____ ~ ____ ~ ______ ~ ____ ~ __________ ~ ____ ~ ____ ~ __________ ~ ____________ ~ __________ ~ ____ ~ __ --------~----~----~~----------~----------~ _____'
o 4 8 12 16 20 24 28 32 36 40 44 48
Time (hrs)
Figure 6.16 Change in Temperature and Bath Depth after a Simulated Decrease in the Target Voltage of O. 1V
25.00 1 .1 L
.
.c
0.
o
o
..
.c
·
"
15.00 tl______
o ~ ________________________
4 8
~ ______
12
=_----~----::----~----~~----~----_;:_--------__~~----~----:;----------~:_----~----~;_----~----::~--~----~
16 20 24 28 32 36 40 44 48
Time (hrs)
10.00 rr- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ,
D•
"•
.J
·
"
Q.
5. oo,L~_~~_~~_~~____=__::__~________::__~_::_~_:;_~___;;_~___:~~____::;__~____::~~_:,
o 4 8 12 16 20 24 28 32 36 40 44 48
T1me (hrs)
990
·..,
L
L
·•
0.
·
E
960L
I ________________________ ~ __________________ ~ ______ ~ __________________ ~ ____ ~ ______________ ----------~----~~----~------------------~----~----------~
o A 8 12 16 20 24 28 32 36 40 44 48
T1me (h,.. .. )
Figure 6.17 Change in Temperature and 8ath Depth after a Simulated Increase in the Target Voltage of O.2V
In the simulation of the step change in target voltage to 4.49 V the bath
depth initially drops as the anodes are raised in the bath then increases as
the ledges and crust melt into the electrolyte due to the overall rise in
temperature associated with a voltage increase.
The effect of anode movement upon bath depth is shown in figure 6.18
which shows the variation in depth due to the large anode movements
during anode effect termination. As expected, lowering the anodes is
associated with a rise in bath depth due to displacement of the electrolyte
by the anode blocks. Returning the anodes to their position prior to the
anode effect retu rns the bath level to its previous depth.
6.6.10 Weight ratio NaFIAIF3 (Ratio)
The weight ratio of sodium fluoride to aluminium fluoride in the electrolyte

is maintained by the periodic addition of aluminium fluoride to the melt and
by mixing fluorides trapped in the exhaust gases with the alumina feed.
The ratio varies due to losses of aluminium fluoride by volatilisation, by
reaction during anode effects to form carbon fluoride compounds and by
reaction with sodium oxide added in alumina feed. The results of
simulation are shown in figure 6.5 for an addition of 25 kg every 24 hours.
This is a similar feed rate to that used at Anglesey Aluminium and
maintains the ratio around a stable average.
The effect of temperature upon ratio is shown in figure 6.19. An increase

in temperature causes an increase in ratio to a new average. Over the first
13 hours this is due to the increase in target voltage from 4.29 V to 4.49 V.
This has caused a rate of increase of about 0.0023 hr1. The results of
bath ratio tests carried out at Anglesey upon three industrial cells are
shown in figures 6.20 to 6.22. In these tests the voltage set point was
increased in two stages by 0.1 volts at each stage. The measurement of
ratio is not continuous and the fluctuations in ratio of the sampling period
makes it difficult to compare the rate of increase in ratio due to the
increase in temperature. The rate of increase of ratio with time is given by
the slope of the ratio graph and ranged between 0.0028 hr1 and
0.0050 hr1. This compares favourably with the simulated value although
the simulated value is at the lower end of the range. The reason for this is
the lower temperature increase in the simulation in which the 0.2 V
185
28.00
E
u
.,
.r=
c.
<II
o
.,
.r=
'"
co
1 8 . 001~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____ ~ ______ ~ ____- L_ _ _ _ _ _ ~ _ _ _ _~
40.0 40 .1 40 .2 40.3 40 . 4 40.5 40 . 6 40.7 40.8 40.9 41. 0

Time (hrs)
8.00
I
.,'"
'0
>
t ~
3.00 1L-____L __ _ _ _ ~
v
_ _ _ _~_ _ _ _~_ _ _ _~_ _ _ _~_ _ _ _ _ _~_ _ _ _L __ _ _ _~_ _ _ _~_ _ _ __ L_ _ _ _~_ _ _ _~_ _ _ _~_ _ _ _ _ _~_ _ _ _L __ _ _ _~_ _ _ _~_ _ _ __ L_ _ _ __J
40 0 40.1 40 .2 40 .3 40 . 4 40 .5 40 .6 40.7 40 .8 40 .9 41.0

Time (hrs)
Figure 6.18 Simulated Effect of Anode Movement During Anode Effect Termination
1. 25
o
.
+"'
'"
+"'
os
1 . 20 L'____L __ __ L_ _ _ _L __ _ ~ _ _~~_ _~_ _~_ _ _ _~_ _~_ _ _ _~_ _~_ _ _ _~_ _~_ _ _ _~_ _~_ _ _ _L __ __ L_ _ _ _~_ _~_ _~_ _ _ _~_ _~_ _ _ _~_ _~_ _ _ _~_ _~_ _ _ _~_ __ L_ _ _ _L __ _~
0.0 1.6 3.2 4 .8 6.4 8.0 9.6 11. 2 12.8 14.4 16 . 0 17 .6 19. 2 20.8 22.4 24. 0
Time ( hr s)
990
'"'-
.."
+"'
'-
'a.E"
'"
I-
---"-~~~~~~----'----;;-~~~~~-'---~~::-~~--'---'------'-------'-----'-----'--
i
960:-
o 0 1.6 3.2 4.8 6.4 8.0 9.6 11.2 12 . 8 14 . 4 16 . 0 17 .6 19.2 20.8 22.4 24.0
Time (hrs)
Figure 6. 19 Simulated Dependence of Ratio upon Temperature

1.3
0
+:;
ctS
a:
1.25 r
~ / \ ~ - -
_ - - -
/ , --
- - 510:=0.0028 Ihr
,,0.0041 (O C
-
~
ctS
en
1.2
1.15
1.1
6' 1000
-
0
(]) 980
-
"-
::J
ctS
"-
(])
a. 960
--
E
~ 940
1000
....-..
~
~ 900
---.....
~
::::l
c. 800
c:::
.....
rn
ANODE
Q) 700 CHANGE
vsp'*"-· .. • • • • • • • •
VSP VSP AE ~.2 Voit8
I
ITt AE .1 .VoIt8
TAP AE
+0.1 .VoIt8
TAP
600
0 10 20 30 40 50 60
Time (hours)
Figure 6.20 Measured Ratio in Anglesey Aluminiums Cell 1861

1.3
- -- -- --
1.25 f ~-----
-
-- -- --
0
:;::;
~
1.2 r slope = 0.0033 Ihr
II:
1.15
-- " 0.00861° C
1.1
0 10 20 30 40 50 60
-
()
0
1000
....
Q) 980
--....
:::J
(Ij
Q)
960
I ~ ~~ "'-/ slope = 0.38° Clhr
a.
E
~
940
0 10 20 30 40 50 60
-
I
1000
I It I 1270 kWh
~
----
.::&.
:::J
900
a.
c 800 ~ vsp
+0.1 Volts
--
(Ij
Q)
700
Tap,AE
----...----
..
Anode Change
-
.
Tap
- --VSp
.
+0.1 Volts
vsp
~.2 v~ .. TAP,AE
I
600
0 10 20 30 40 50 60
Time (hours)

1.3
1.25
0
:;::; ~ = 0.0053
~
II:
1.2 r ~ ... ...
...... slope Ihr
" 0.0088 I" C
1.15
1.1
0 10 20 30 40 50 60
-
()
0
1000
0) 980
'-
:::J
......
'-
C13
I c- ..G---...c.
~~---V slope = 0.6
0
C/hr
O)
a. 960
E
~
940
a 10 20 30 40 50 60
-
I
1000
~
-
~
......
:::J
a.
900
800
c
......
VSP 1\ ;\ ~2
vSP
Volt
C13 700 TAP -H).I Voila

AE - TAP.AE v~,...
0) AE AE TAP
I
600
a 10 20 30 40 50 60
Time (hours)

increase is accompanied by a rise of 6 °C compared to a rise of about
15°C in the experimental results.
In the simulation the ratio has increased by about 0.03 for a temperature
rise of 6 °C giving the temperature dependence of ratio as 0.005 °C-1. The
Anglesey data shows some agreement with this, with the temperature
dependence of ratio in figures 6.20 to 6.22 ranging between 0.005 °C-1
and 0.0075 °C-1.
6.6.11 Temperature
Typical temperature time curves for two different feed frequencies are
shown in figure 6.23[61. The temperature is measured at the tap end and
due to imperfect mixing in the cell the effect of tap end feed is greater than
that for duct end feeding. In order to compare this to the simulation which
assumes the electrolyte to be thermally well mixed, the tap and duct ends
of the cell will be assumed to be independent of each other. The graph
can then be compared with simulation of about a 70 minute feed time.
The results of simulation for a feed frequency of 90 kg every 67 minutes

are shown in figure 6.24. This shows a rise of temperature of some 8 °C
over the feed cycle which compares well with the 11°C rise in the
industrial data. The time over which the temperature falls due to an
alumina feed is slightly less in the simulation at 9 minutes compared to 15
minutes. This is because the model assumes that any chemical addition to
the electrolyte is instantly raised to the temperature of the melt. In
practise, the rate of heat transfer to the alumina particles is limited by
agglomeration of particles and the formation of sludge. The overall heat
required to raise the temperature of the alumina and hence the
temperature decrease associated with a feed is unaffected by the rate of
heat transfer. As the simulated cooling time is of the same order as the
experimental data the use of this simpler model is justified.
The effect of a step change in voltage upon the average temperature in

cell is shown in figure 6.25[61. This is industrial data for a change in cell
voltage of 0.1 V at time 20 hours with temperature being continuously
measured at the tap end of the cell. The average temperature of the cell
increases from 976°C to 981 °C over a period of approximately 2 hours,
giving a rate of temperature rise of 2.5 °C hr1. The results of simulation of
186
9
985i I FEED TAP END

I FEED DUCT END
/FEED TAP END
I
0
0
I
W 980
0::
:;)
~
~ \FEED DUCT
0:: END
~ •
w
.-
975
\
~
4
~
I
I \.V /
~I\"
V TYPI CAL TEMPERA TURE -

TI ME CURVES FOR TWO
965~ l /' 01 FFERENT FEED MODES
(Al '" 37 MI N BETWEEN FEEDS
(B)" 3 HR. BETWEEN FEEDS
960
I
0 20 40 60 80 I~ 120 140 160 180 200
TIME - MINUTES
Figure 6.23 Typical Temperature- Time Curves for Two Different Feed Modesf6)
o
N
~ ~
~
~
(.)
'" '" ~
Q)
~
~
:::J
~I'-..
~ ~
(()
~
~
~
~
0
Q)
~
0
"'
L-
.<:
~
"'
L-
.<:
e
a..
QI QI
E
E
.... >= @
:::J
......
~
Q)
CD CD Q..
0 0
E
~
~
Q)
......
'" ~
0
~
0
:::J
.§
CI)
~
~
C\J
~ 0 (()
0
@
:::J
.~
l(
N
N
0
,;
o L __- L__- L__- L__- L__- L__-U~-L---L--~--~ ~

o 0
o o 0 o o
o
CD
'" '"'" COZ LV %l/l

'"
~ ~ . ' :." t
% "
Z 0 LIJ
~~
-' w Wu~
-' 0 OWl
0 LIJ LIJ LIJ
::f~ ~
LIJ W OI.4..W LIJ
~ z 0 Z~O 0 0 I-
wI- ~ ~ ex ~ ~W~ ~ ~ ~ "-
o(/)
W W W W W W LIJ Cl.:) W W
~
"-
0
~
~
"-
o
~
l-
~ e "- ~
~5
0
~ ::1
w
a::
:J
~
ex
a:: 97
w
a..
~
w
~
~
t·37
ex
CD
CELL VOL. TS INCREASED - 0.1
BA TH TEMP. - 976°C A BATH TEMP.- 981°C AVG.
12 13 \4 15 16 17 18 19 20 21 22 23 24 2 3 8 9
HOUR OF DAY
Figure 6.25 Average Temperature in an Industrial Cell After a O.tV Increase in Operating Voltage/61
a 0.2 V increase in target voltage upon the temperature and ratio of the
electrolyte are shown in figure 6.19. It is difficult to judge the rate of
increase of temperature from this graph due to the fluctuation throughout
a feed cycle but the overall increase in average temperature is about 6 0C
from 968°C to 974 °C. The accompanying increase in ratio occurs over
the first 13 hours of simulation which suggests that this is the length of
time that the cell takes to reach its new average temperature. Based upon
this, the temperature rise is 3 °C / 0.1 V at a rate of 0.46 °C hr1.
The temperature data for Anglesey's cells shown in figures 6.20 to 6.22
was measured at 2 hourly intervals and it is therefore difficult to determine
an accurate rate of increase of temperature due to changes in cell voltage.
The overall rate of temperature rise for an increase of 0.2 V is in the range
0.38-0.67 °C hr1 based upon the time taken for the cell to achieve a new
steady temperature. The overall temperature rise, however, is about
7.5 °C / 0.1 V.
The differences between the two sets of industrial data reflect the different
dynamic behaviour of different designs of reduction cell. There is
agreement, however, in the overall temperature increase with increase in
cell voltage. The temperature rise in simulation is somewhat less, being
approximately half that found in commercial cells at 3 °C / 0.1 V.
It is difficult to accurately determine the reasons for this discrepancy as

the information regarding the industrial measurements is incomplete. The
average current in the industrial cells is similar to that in the simulation and
the increase in power input to the cell is therefore the same in the both
cases. The net increase in power input however also depends upon the
change in heat loss from the cell, the change in heat transfer to the frozen
ledges and any change in the rates of chemical reactions.
(6)
The temperature increase in the cell is then determined by dividing the net
power input by the thermal mass of the cell
dT ~QI.net (7)
-
dt M bath . CPbath + Mpad . CP pad
187
where Mbath = mass of electrolyte
CPbath = specific heat of electrolyte
Mpad = mass of aluminium pad
CP pad = specific heat capacity of aluminium pad
The main parameters affecting the temperature increase are therefore the
volumes of aluminium and electrolyte in the cell and the composition of the
electrolyte. These directly affect the thermal mass of the cell and
contribute to the heat transfer and heat loss terms in the determination of
the increase in net power input.
The heat transfer to the sidewalls is given by
(8)
(9)
where h = heat transfer coefficient

Tb = bulk temperature of electrolyte and metal pad
T 1iq = liquidus temperature of the electrolyte
suffix
bf refers to bath/ledge interface
mf refers to metal pad/ledge interface
The area for heat transfer is directly related to the depth of the electrolyte
and aluminium pad and hence the volume of these two sections. Both the
heat transfer coefficient and liquidus temperature of the electrolyte are
dependent upon the composition of the electrolyte. As these parameters
are not available for the industrial cells it is impossible to determine the
accuracy of the temperature rise predicted although it is to be expected
that the Anglesey Aluminium cells are operated at a composition close to
that used in the model. It is therefore likely that the heat transfer
coefficients used in the model require some tuning with the simulated
results being compared to additional industrial measurements where the
above parameters have been recorded.
188
A contributing factor to this discrepancy is also the increase in current
efficiency in the simulation causing a reduction in the heat input to the cell.
In practise, an increase in temperature is usually accompanied by a
reduction in current efficiency caused by an increase in the rate of the
back reaction of aluminium to alumina and an associated increase in heat
input. In the simulation the increase in the interelectrode gap to acheive
the new target voltage appears to give a greater increase in current
efficiency than the decrease accorded by the rise in temperature. This
reduction in the rate of the back reaction reduces the heat output of the
chemical reaction and hence the net heat input. It may be that the model
for the dissolution of aluminium into the electrolyte has too great a
dependency upon the interelectrode gap and not enough upon the
temperature and this should be investigated further.
A continuous record of electrolyte temperature during anode effect in a

commercial cell is shown in figure 6.26[6]. This shows an anode effect
lasting about 4 minutes. The normal increase in bath temperature
accelerated markedly about six minutes prior to anode effect. During
anode effect, the temperature increased rapidly reaching 1020 °C before
the anode effect was terminated. The temperature returned to the value
prior to anode effect in about 10 minutes due to the addition of alumina
feed.
The simulated variation of temperature during anode effect is shown in

figure 6.27. This shows good agreement with the data in figure 6.26 with a
marked increase in the rate of temperature rise about five minutes prior to
anode effect and the return to the pre-anode effect temperature about 10
minutes after the anode effect termination sequence has been activated.
The maximum temperature achieved in the simulation is 30°C less at
990°C associated with a maximum voltage during the anode effect of
27 V. Unfortunately, no information is given for the maximum voltage
achieved in the commercial cell, although this is typically in the range 25-
40 V and it is reasonable to assume that the higher temperature is due to
a larger anode effect voltage.
189
t-
o
W
Ll..
Ll..
W
W
1030 o
o
z
c:t
1020
;010'
100
o
w
W
Ll..
980
t
97Gl -
35 30 25 20 15 10 5 10 15 20 20 30 3"5 40
MIN BEFORE AE AE MIN AFTER AE
Figure 6.26 Temperature in an Industrial Cell During Anode Effectf6}

1000 r------------------------------------------------------------------------------------------------------------------------------------------,
u
...
.....
.,
::I
..
c.
E
>-
.,
..
.c
<D
960~
1 ____ ~ __ ~ ____ ~ __ ~ __ ~~ __ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ __ ~~ __ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ ____ ~ __ ~ __ ~~ __ ~
o 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
Time (min)
30.00
[G
...'"
o
>
O. OOLI__~____~__~__~__~__~____~__~__~__~__~__~~__~__~__~__~__~____~__~__~__~--~--~~--~--~--~--~--~----~~
o 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60
T1me (m1n)
Figure 6.27 Simulated Temperature Profile During Anode Effect

6.7 Plant Trials
Since this work was completed at Newcastle University, the simulation has
been in use at Anglesey Aluminium to investigate the cell operation. An
early study considered the effect of changing the metal tap interval upon
the temperature and current efficiency of the cell. The results of the
simulation study are shown in figures 6.28 to 6.30.
It was unfortunate that the tap at around 50 hours coincided with the cell
being in search mode. The automatic tapping routines are unable to
successfully restore the cell voltage to its target value under these
conditions and this leads to an artificially long period of high voltage
following the tap. The average cell conditions are distorted by this, but
comparison of the tapping regimes is still valid.
As expected, the simulation shows a slight decrease in the cell

temperature from the 24 to 32 hour tapping regimes. This is primarily due
to the increased heat transfer from the cell through the increased metal
pad / frozen sidewall interface. As was discussed in the previous section,
the simulation appears to be a little unresponsive to changes affecting the
cell temperature and it may be that improved tuning of the model would
enhance this effect.
The 48 hour regime shows no further improvement in either the

temperature or the current efficiency, the efficiency being slightly improved
over the 24 hour tap regime. An interesting feature of the current
efficiency curves is the difference in shape of the curves between feed
times, In the first fifty hours, prior to the period of high voltage and
temperature, the rate of increase of current efficiency increases rapidly as
the alumina composition is depleted at the end of the feed cycle. In the
second half of the simulation, where the temperature is higher, the current
efficiency curve shows a much more rounded profile, with the rate of
increase of current efficiency tailing off towards the end of the feed cycle.
This smoother profile leads to a greater average efficiency over the feed
time and the unexpected result of increased current efficiency with
increasing temperature. This is due to the strong dependence in the
model of the aluminium dissolution on the interelectrode gap. In should be
noted that in the second half of this simulation the acd is unreasonably
large and as such may not be truly indicative of the performance of the
190
100
en 90
fa
<D
C\I
N
OJ 80
>-
(,)
cQ)
'0
!E 70
UJ
C
~
~
() 60
50
o 10 20 30 40 50 60 70 80 90 100
30
~
en
fa
E
~ 25
oC\I
o/j
E
~
lli
,....
20
15
10
o 10 20 30 40 50 60 70 80 90 100
990
985
en
>
co
<D 980
o
"-
OJ
975
~
~
"§ 970
Q)
Cl.
E 965
~
£
iii
CD
960
955
950
70 80 90 100
o 10 20 30 40 50 60
Figure 6.28 Simulated Results of 24 Hour Tapping Schedule

100
01 90
>
OJ
CD
C\J
(\j
Ol 80
>-
()
c
Q)
'0
:E 70
w
C
~
::::l
0 60
50
0 10 20 30 40 50 60 70 80 90 100
30
01
>
OJ
E
()
-=t 25
ci
C\J
0<:1
E Bath depth
()
-=t
o.ri 20
,....
.c
Cil
CD
u Metal pad
c 15
OJ
~
Q)
Tap Tap Tap
~
10
0 10 20 30 40 50 60 70 80 90 100
990
01
iO 980
CD
ci
f'..
Ol
~ 970
::::l
~
Q)
Cl..
E
~
.c 960
Cil
CD
950
o 10 20 30 40 50 60 70 80 90 100

100
90
dl
~
C'?
N
en 80
()
c
Q)
T5
~ 70
w
C
~
~
0
60
50
o 10 20 30 40 50 60 70 80 90 100
30
dl
~
E 25
0
v
ci
N
o/l
§
N
r--: 20
.J::.
11i Metal pad
CD
"0
c 15
C1l
Tap Tap
~
::2
10
0 10 20 30 40 50 60 70 80 90 100
990
01 980
>
C1l
<D
0)
<D
en
~ 970
~
~
Q)
a.
E
~
.J::.
11i 960
CD
950
70 80 90 100
o 10 20 30 40 50 60

model. This is however, further evidence that the current efficiency in the
model may be biased towards acd rather than temperature and more work
will be required to establish whether this should be the case.
6.8 Conclusions
It can be concluded from the results of this work that the simulation
successfully simulates both the behaviour of the cell and the control
system, although further tuning of the heat transfer parameters would
seem to be indicated. It would also be advisable to further investigate the
dependence of the current efficiency upon the cell temperature as this
does not seem to be sensitive enough in the current model. The modular
nature of the program makes the changing of individual algorithms
straightforward and facilities have been included to allow the integration of
additional user menus an easy process.
The simulation can now be used to investigate different control systems,

both by changing the automatic control algorithms in GELTROl and by
changing operating procedure as in the tapping regime example. The
modular structure of the program makes it a very powerful tool for control
system analysis, as new control strategies may be developed in isolation
and simply linked into the model of the cell. This is analogous with the
implementation of new GELTROl algorithms on site.
In addition to the feedback control of 'conventional' parameters, for

example voltage, the simualtion may be used to investigate the
possibilities of controlling the cell based upon information that may not be
currently available in the commercial cells. For example, a control system
could be developed using the feedback temperature and alumina
concentration. Whilst it may not be possible to duplicate this system in the
plant the simulation could be used to obtain information regarding the
target parameters for the commercial cells and could indicate whether the
development of continuous measurement methods for these parameters
would be worthwhile.
191
al6c3 al6c4 al6c5 al6c6 al6c7
Current (kA) 151.0 151.0 151.0 151.0 151.0
Cell Voltage (V) 4.40 4.30 4.50 4.41 4.41
Target Volts (V) 4.39 4.29 4.49 4.39 4.39
Bath Depth (cm) 17.95 18.68 20.49 18.94 19.26
Pad Depth (cm) 13.26 12.89 12.72 14.18 11.86
Pad Ledge (cm) 6.64 7.45 6.24 6.72 6.63
Current Efficiency (%) 90.46 89.73 91.22 90.44 90.48
ACD (cm) 4.67 4.36 5.00 4.68 4.69
Alumina Conc. (wt%) 3.38 3.51 3.10 3.29 3.20
Bath Temp °C 970.5 968.6 974.2 971.4 972.5
Liquidus Temp °C 959.6 958.6 962.6 960.7 961.4
Wt Ratio 1.231 1.221 1.240 1.234 1.237
Table 6.3 Average Parameters From Five 170 Hour Simulations of a Half
Break Cell
References
1 Anglesey Aluminium Internal Memorandum
2 Waddington, J. and Pearson, T. G., "Electrode Reactions in the

Aluminium Reduction Cell", Discussions, Faraday Society, 1947, vol 1,
pp307-320
3 Tabereaux, A.T. and Richards, N.E., "An Improved Alumina

Concentration Meter", AIME Light Metals, 1983, pp495-506
4 Wilson, C.A and Tabereaux, A.T., ''Alumina Control in Center-Break

Cells", AIME Light Metals, 1983, pp479-493

6 Dow, D.W. and Goodnow, W.H., "Influence of Operating Variables on

Reduction Cell Bath Temperature", AIME Light Metals, 1972, pp3-20
192
7.0 Recommendations For Further Work
7.1 Inferential Control
An area of work worthy of investigation with the model is real time control
of the aluminium cells using the model to predict parameters in the cell
that cannot be measured or are not available on a continuous basis, for
example, the alumina concentration. By connecting a real time version of
the simulation in parallel with the cell as shown in figure 7.1 the simulation
could output the required control signals based upon continuous results of
simulation. This would allow the optimal control strategy to be developed
continuously for any particular cell.
The model would continually correct its predicted parameters based upon
any available data from the cell, eg voltage, tap time, length of anode
change, periodic temperatures and electrolyte composition.
The first stage in this work was to develop a real time version of the
simulation that was controlled by a GELTROl panel. This was then
connected in parallel with a working cell with the aim being the
comparison of available experimental data against the predicted
parameters. This open loop system would allow the tuning of the model to
the particular cell aid in the validation of the model.
7.2 Development of the Real Time Simulation (RTS)
In the development of a Real Time Simulation three areas have to be

addressed:
1. A real time kernel

2. Data logging interface
3. Data storage / Operator interface
1. In order for the model to be used successfully for real time control, the
simulation must run several times faster than real time. In order to co-
ordinate the model and the logging of experimental data for comparison a
real time kernel must be developed. The kernel has a number of functions
as shown in figure 7.2.
193
Voltage
Fe edtime
Cell Model
Feeds
Figure 7. 1 Real Time Simulation in Parallel with Working Cell

Model
Operator
Interface
Screen I Keyboard
Datalogging Ioo-o------t
IControl
Aluminium Cell Central timing

Data
Storage
Disk I Printer
Figure 7.2 Use of Model in Real Time Control
The real time kernel must schedule the events depending upon a pre-set
timing and priority. In the simulation the order priority of the tasks is
1) Data logging / Control The control system operates on a 0.1 s interval.

If a control signal, eg the raiSing of an anode, is missed then there will
be poor agreement between the model and reality. Data logging takes
a small amount of run time and therefore can be made a high priority,
without being detrimental to the running of the model.
2) Model For good agreement between the model and reality the
process must be simulated at a frequent interval. Additionally, it will be
necessary to run the model a number of times adjusting various
parameters until the results agree with the logged data. The
simulation of the process must therefore have a high priority and must
also be fast. It may not be necessary however to solve the equations
faster than the base interval for the control signals. The noise in the
measured data, eg Voltage, prohibits the use of raw data logged at
the same interval as the control signals and a filtered signal should be
used. It would be sufficient to compare the 'results' at intervals of say
30s, so long as all the control actions are known.
3) Data Storage Data storage to a permanent medium is a comparatively

slow process and longer intervals between data sets are generally
required. There is no need for the data to be stored all at once but
instead it can be buffered and stored during periods when the
simulation is in an idle state.
4) Operator Interface The facility must exist for an operator to

communicate with the model without interrupting the running of the
simulation. This will allow the model to receive information regarding
the process that is not available through instrumentation and allows
the operator to change the display and storage parameters. This task
has a low priority as it must not be allowed to interrupt the simulation
or the data logging.
A very simple real time kernel can consist of a timing loop that calls each
of the tasks in turn , where the time taken for each task is known. If the
lower priority tasks are known to take longer than the interval required for
194
the higher priority tasks then the higher priority tasks may also be called
by the lower priority tasks. For example, if there are three tasks
DATALOGGING, MODEL and STORAGE with descending priorities the~
the kernel could be written
Call STORAGE
Timer loop
STORAGE: Call MODEL
Call DATALOGGING
Call MODEL
Call DATALOGGING
MODEL:
Call DATALOGGING
Call DATALOGGING
DATALOGGING: ...
where the ellipsis represents the required code for each task.
This method is cumbersome and requires a great deal of tuning of the

timing loops to ensure that all tasks are complete within the required
intervals. It does not, however, require any multitasking capabilities and is
applicable to most machines and languages.
This approach was originally tried in order to validate the response of the
model to the control actions of the CELTROL computer. The program was
written in PASCAL and connected to the CELTROL panel via a TeS
Signal processor.
195
The CElTROl panel outputs digital signals for the control of the raising
and lowering of the crust breakers, the operation of the feed dump and the
raising and lowering of the anodes. It inputs an analog signal of the cell
voltage in the range 0 to 30 V. The CElTROl 10 signals are represented
schematically in figure 7.3.
The simulation was connected via the TCS signal processor to the
CElTROl panel. Communication between then signal processor and the
computer was via the RS232 serial port. This is shown schematically in
figure 7.4.
The simulation outputs the simulated cell voltage to the CElTROl panel
which returned the control signals for the simulation. Initial trials with an
isolated CElTROl panel were successful with the simulation responding
correctly to the signals from the panel to raise/lower anodes and feed the
cell. The next stage was to connect the simulation in parallel with a
working cell so that the simulation would receive the same control signals
as the cell in order that the results from the simulation might be compared
with the data from the cell. This configuration is shown in figure 7.5.
In this case the CElTROl computer operated upon the actual cell
voltage. It was also necessary for the simulation to use the actual line
current that was being supplied to the cell. This was obtained as an
analog signal through the TCS signal processor.
This experiment was unsuccessful because of the communication link

between the computer and the TCS processor. The use of an RS232 link
to the signal processor did not allow a reliable means of handshaking
between the devices at the speed required and it was necessary to
communicate at a low baud rate of 1200 baud. At these speeds however,
the simulation was unable to run sufficiently fast to complete the data
logging and model solving tasks in the required intervals. The failure to
communicate properly with the signal processor caused the computer to
hang and it was impossible to get any useful data from this trial, using a
very simple real time kernel.
A more sophisticated real time kernel can be written using languages that
allow concurrency, that is allow multiple tasks to be run simultaneously.
196
R'
alse A..nodes
Lower An odes
...
Line Current
... Crust Breaker (Tap End)
...
CELTROL Crust Breaker (Duct End)
...
...
Dump (1ap End)
Ce II Voltage ......
Dump (D uct End)
......
Figure 7.3 GEL TROL Ana/og/Digita//nputs and Outputs

Line Current
mrs
CELTROL
Link Control Actions (Digital)
Real Time AID Signal

Simulation Processor
Figure 7.4 RS232 Link Between Simulation and CELTROL

Line Current
Control ActIons (DIgital)

ID
Real Time AID Signal

Simulation Processor Working Cell
Figure 7.5 Simulation Connected to Working Cell

True concurrency can only be achieved by using hardware that allows
several tasks to be processed simultaneously, although pseudo-
concurrency may be achieved simply through software.
The Modula-2 language running under MS DOS allows pseudo

concurrency and a simple real time kernel may be written to manage the
required tasks. MS DOS is, however, non-re-entrant. That is, if DOS is
processing a command and control is transferred out of the DOS
command to another process, it is not possible to return to the DOS
command. This appears to be a problem when writing a real time kernel
but may be avoided by observing the basic principle of avoiding task
switching whilst DOS is in a critical section. This can be determined by
calling the DOS interrupt 21 H function 34H which indicates if DOS should
not be interrupted.
The flow diagram for the scheduler is shown in figure 7.6. The scheduler
should be assigned to the timer interrupt DOS interrupt 1CH which is
called 1024 times every second. The timers should be set to the number
of timer ticks required between each call to each process and the timer
tests must be performed in order of priority. The lowest priority task may
be run continuously as the main process to be interrupted. In the
aluminium simulation, the operator interface and data storage may be run
as the main tasks with the high priority tasks of data logging and the
model being handled by the scheduler.
A disadvantage of relying upon DOS not being in a critical period is that it

reduces the times when control may be transferred between processes. It
is therefore necessary to reduce the number and frequency of critical DOS
calls. The main DOS processes that cannot be re-entered are the 10
routines, writing to the screen, input from the keyboard and data storage
on disk. This may be avoided by directly accessing the relevant hardware
without using the DOS functions. This does however make the program
very machine specific. The modular nature of Modula-2 can be used to
great advantage here. Although the low level routines such as screen I/O
must be machine specific, these can be stored in modules with the same
names and means of use on all machines. It then becomes very easy to
transfer a program between machines by simply recompiling the software
with the relevant modules.
197
Return to
decrement timers current process
no
no
.....-
yes yes yes

yes
Return to
current process
is
DOS in
critical
\ . ------
section?
no
Reset timer P
Transfer to
process P
Figure 7.6 Flow Diagram for DOS Task Scheduler

Time did not allow for any more than preliminary work to be carried out on
a real time simulation of the aluminium smelter. An elementary real time
kernel was developed to schedule two simple liD tasks writing to the
screen and taking input from the keyboard. This was successful and would
form the basis of a more advanced kernel allowing the scheduling of a
greater number of tasks.
7.3 Tuning the Model for Control
Having developed a successful real time simulation of the process, the

next stage would be to tune the model to give accurate responses to
disturbances to the process and to develop a method by which the model
could adjust the stochastic parameters to achieve good agreement with
the process. The model could then be used to provide a much greater
level of control of the process.
In the model a much greater variety of variables are available for control,
most of which it is not possible to measure continuously on the process.
An example is the control of alumina feed based upon the concentration of
alumina in the cell. At present it is not possible to continuously measure
the alumina concentration in the melt and feed back control methods to
the alumina feed time are based upon the rate of rise of cell voltage as it
approaches anode effect. If the alumina concentration was known,
however, this could be directly fed back to the alumina feed algorithm to
control the alumina concentration at much more precise levels.
This use of the model is shown in figure 7.7. The simulated alumina
concentration is fed back to the control system to determine the amount of
alumina to be fed. The control signal is then passed to both the model and
the process. All available measurable parameters from the process, in this
example, the cell voltage and temperature, are then compared with the
predicted values from the model. Discrepancies will arise between these
parameters due to uncertainties in the control action. In this example,
there is a variation on the amount of alumina fed to the cell. This
discrepancy is used to alter effects of the control action in the model until
the difference between the measured and predicted variables is within the
required tolerance. The simulation is then continued and this procedure is
repeated at the next integration interval.
198
, I'\.
predicted
h• [A1 20 31
. £ U
-, l: controller l: model correction to
mass of
T, V
AI20 3 fed
comparator/
corrector
~
T, V
process
Figure 7.7 Inferential Control

The complexity of the control system in the model is not limited to practical
considerations such as the problems of monitoring variables in a hostile
environment and it is therefore possible to control the process upon any
variable that has been included in the model. A complex control system,
however, that included a number of simultaneous control actions would
require a complex comparator capable of distinguishing between the
effects due to stochastic disturbances to the different control actions.
7.4 Alumina Concentration Meter
The importance of the alumina concentration In the operation of the

aluminium cell has made the development of a direct and continuous
means of measurement highly desirable. An alumina concentration meter
was first patented by the Reynolds Metal Company in 1969[1] and the
method was improved by Johnston and Richards in 1974[2] and by
Tabereaux and Richards in 1983[31.
The alumina concentration meter employs the electrochemical relationship

between the alumina concentration in the electrolyte and the limiting
current density for the oxide reaction at the anode. As the cell approaches
anode effect, the oxide reaction becomes mass transfer limited and the
partial current density for the reaction becomes equal to
(1 )
where 10 = partial current density for the oxidation of oxygen

Z = charge transfer number for the reaction =2
kL = Mass transfer coefficient
[0 2-]b = concentration of oxide ions in the electrolyte
= concentration of alumina
The limiting current density can be determined by monitoring the voltage

across the cell as the current density is increased. When the current
density at the anode reaches the limiting current density for the oxygen
containing anions a rapid increase in voltage will occur. This relationship is
shown schematically in figure 7.8.
199
~
·s
E
~
«
1:)
c:
~
~ .0
r--"'I
M '-
(J)
c:
0C\I Q)
a......
-
« c:
........ W ~
10....:
:J
... ()
-CO c::
0) 0
:§ '-............
--
+-' ~
c: .§ c::
Q)
.....J
Q) (.)
+-' c:
Q)
c:
0
0
a.. ~
......
()
Q)
co
.9-
.c:
(J)
c:
'- 0
......
~
Q)
E
._1 cc:
-..;::-- co
1
1
t-..:
~
:J
.0)
Li:
lUaJJn8
- 'Al!SUaa
I
The alumina concentration meter comprises a probe, that inserts into the
molten electrolyte, and a voltage-current programmer that measures the
voltage drop across the probe for a programmed series of imposed current
steps. The probe is shown in figure 7.9 and consists of graphite anode
and cathode separated with a boron nitride insulator encased in a steel
sheath.
After calibration in the laboratory, using synthesised bath and a range of

specific alumina levels, Tabereaux and Richards immersed the probe in
the bath of a 100 kA Soderberg cell for a 42 hour period and compared
the concentration measurements obtained with the probe against those
obtained by chemical analysis. The alumina profile obtained with the
probe were similar to that obtained by chemical analysis indicating that
this technique is generally able to correctly monitor the variation of
alumina in aluminium reduction cells. The values obtained by the meter
were generally within 0.25 wt°jo of those obtained by chemical analysis
although there was a tendency for the meter to read high after alternate
feeds to the cell which was probably due to localised conditions around
the probe.
Although the alumina concentration meter appears to be reasonably

accurate as a direct measurement device, it has a number of drawbacks
when considered as a means of continuously measuring the alumina
concentration for feedback control. The hostile environment of the
reduction cell is unfavourable to a long service life of this type of
instrument and the installation of a probe for every cell in a pot room
would be cost prohibitive. Additionally, Tabereaux and Richards indicate
that the probe tip required fairly frequent resurfacing to obtain reproducible
results and this would be undesirable in a working cell. Finally, the effects
of local conditions upon the results obtained by the meter would require
that more than one probe be installed in each cell to provide a reliable
indication of the general conditions in the cell and this would again be cost
prohibitive.
In order to utilise this technique for the direct measurement of alumina

concentration it will be necessary to devise a much simpler probe that can
be manufactured and installed cheaply and easily within the cell. The
approach taken was to incorporate the measurement anode within the
200
+
-
Graphite Anode
and Cathode
Steel Tube
Boron Nitride
Insulator
Figure 7.9 Electrolyte Probe for Reynolds Alumina Concentration Meter

prebake carbon anodes used in cell as shown in figure 7.10. The
measurement probe consists of a rod of carbon drilled from the anode and
then insulated from the main anode by solid cryolite. A potential control
device is connected across the carbon rod and the cell cathode and this is
used to determine the limiting current density at the anode by forcing the
carbon rod alone to anode effect conditions.
Trials of this device were started at Anglesey Aluminium in a half break

cell, with the probe being drilled from the anode using the existing drill
used for taking quality control core samples from the anode. The core was
approximately two inches in diameter with the drill leaving a hole with a
diameter around three inches. Unfortunately, the drill was in a fixed
position above the anode conveyor belt and it was not possible to fit a
long enough bit to drill completely through the anode. In order to obtain a
continuous rod from the anode it was therefore necessary to drill the
sample from both sides with careful alignment of the drill. This task was
extremely difficult and in the time available for the trial only a single rod
was successfully obtained.
The rod was inserted back in the anode, and the gap between the rod and
the anode was packed with crushed cryolite. The anode was then loaded
into the cell normally and the connections made to the potentiostat device.
Although installation appeared to have been successful, the trial did not
yield any satisfactory results and the anode was removed early due to
poor performance. It was impossible to determine whether the anode
failure was due to the insertion of the probe or simply due to the anode
itself.
Although this trial was unsuccessful, it is recommended that this research

be pursued further. This design for an in-situ alumina concentration meter
has a number of advantages over other types of probe:
1. The probe is relatively easy and cheap to manufacture, allowing it to

be installed in multiple positions within the cell. This could allow a
better estimate of the true bulk concentration than that obtained from
a single measurement. The manufacture of the device could be
simplified by using a drill designed for the task. It may also be
201
CELTROL
Potential
Control
Device
/ / / /
/ / / /
/ / / /
/ / / / / / / / / / / / /
Molten Electrolyte ,.
/ / / / / / / / / / / / / /
Aluminium Pad
Figure 7. 10 In Situ Alumina Concentration Meter

possible to improve the insulation between the probe and the anode
by using molten rather than solid cryolite.
2. Due to the highly corrosive nature of the electrolyte, probes

containing metal components tend to be dissolved into the bath.
Although this can introduce only small amounts of impurities into the
system, continuous usage is likely to have a significant effect upon
the purity of the aluminium produced in the cell. As the only
components of this meter coming into contact with the bath are the
anode material and the cryolite insulation, there can be no possibility
of contamination.
3. This design could easily be incorporated in the existing distributed

control system for logging the concentration or for use in feedback
control. The only permanent component is the potential control
device, circuitry for which could be incorporated in the GELTROl
control panel. The probes would be routinely replaced at every
anode change although this would necessitate the development of a
simple and robust means of connecting the probe to the control
panel.
References
1 Reynolds Metals Company, "Alumina Concentration Meter", U.S. Patent

3,471,390, October 7, 1969
2 Johnston, T.J. and Richards, N.E., "Laboratory and Field Studies with an
Improved Direct Reading Alumina Concentration Meter", AIME Light
3 Tabereaux, A.T. and Richards, N.E., 'j4n Improved Alumina

Concentration Meter", AIME Light Metals, 1983, p495-506
202
Appendix A Initialisation of the Simulation Program
A.1 Initialisation Files
The aluminium reduction cell simulation program, AlSIM, is initialised

from two ASCII text initialisation files, one containing the initialisation
parameters for the cell model and the other containing the initialisation
parameters for the control system, CElTROL. The file used to initialise the
model parameters in the simulation of the 'standard' run used in this work
is listed below.
At the end of every simulation, the program dumps the model and control
parameters to disk in ASCII format. These 'end ' files can then be used as
the initialisation files for subsequent simulations.
(*This is the end file of a 170 hr run with model Sc. It will be used as
(*an initialisation file for further runs on that model.
**OATA**
(* Model Version
SIXC
(* Cell type
HALF
(* current A
1.5100E+005
(* random number seed absolute variance on feed ± 20%
3 2.0000E-001
(* mass of feed g/dump
2.3000E+004
(* mass of bagged AI F3 added g
2.5000E+004
(* pad depth bath depth acd cm
1.2783E+001 1.7027E+001 4.S014E+000

(* ratio wt% alumina wt% CaF2
1.2211 E+OOO 2.90S4E+000 5.3509E+000
(* Temperatures C
(* bath surrounding cavity crust
9.7203E+002 2.5000E+001 2.0000E+002 4.0104E+002
(* Anode dimensions cm
203
18 8.0000E+001 1.3460E+002 4.0000E+001 2.2S00E+001
S.OOOOE+OOO
1.S600E+000
(* External pot dimensions cm

8.8800E+002 4.0000E+002 4.7800E+001
(* Cathode dimensions cm
8.1000E+002 3.2200E+002 3.8000E+001
(* shell wall thicknesses cm
1.2000E+001 2.2000E+001
(* freeze thicknesses and alumina cover cm
1.237SE+001 6.9079E+000 7.0103E+000 S.OOOOE+OOO
(* htc bath, pad, shell, exposed bath W m-2 K-1
6.0000E-002 1.0000E-001 2.0000E-003 1.0000E-002
(* htc cavity break padS crust anode W m-2 K-1
1.0000E-003 1.0000E-002 1.S000E-001 6.0000E-002 8.0000E-002
(* Thermal conductivity kshell ~reeze kalumina W m- 1 K-1
3.1000E-001 1.0000E-002 4.0000E-003

(* Thermal conductivity kbath kcrust kanode kcathode W m- 1 K-1
2.0000E-002 8.0000E-003 4.4000E-002 8.0000E-002
(* Specific heat capacity cPal cPbath cPalumina J 9- 1
1.0800E+000 1.6S70E+000 2.0000E+000
(* Specific heat capacity cPanode cPcathode CPledge J 9- 1
1.S4S0E+000 1.S4S0E+000 1.6S70E+000

(* Enthalpy of alumina at 473, 1200, 1300 K
1.6400E+004 1.0300E+00S 1.1600E+00S 1.3000E+002
(* Heats of reaction main reaction, back reaction, anode effect reaction
9.9600E+00S -3.S900E+00S 8.0800E+00S
(* Alumina heat of dissolution, latent heat of fusion
1.2S00E+00S S.S300E+002
(* External Resistances
(* Anode bus, riser+flex Ohms
4.1000E-007 3.8000E-007
(* rod+clamp, stub to anode Ohms/anode
1.0440E-00S 1.4S80E-00S
(* cathode, cathode flex, cathode bus Ohms
2.3000E-006 2.0000E-007 1.2400E-006
(* resistivity of anode
204
4.0000E-003 6.9000E-003
(* initial masses
(* alumina in suspn, sludge added by anode change aluminium in solution
6.5510E-003 1.6442E+001 2.2000E+004 1.2924E+003
(* dissolution rate parameters kO tempO
1.2450E-003 1.3830E+003
(* specific areas for alumina in sludge and suspension
7.5000E+001 1.5000E+002
(* sludge factor
5.0000E-001
(* Tafel coefficients for oxide reaction ao bo and anode effect af bf aev
4.7250E-004 6.9500E-005 2.5500E+OOO 3.0000E+001 5.0000E+OOO
(* Anode effect bubble parameters thickness,leakage,conductivity
6.1000E-001 1 .OOOOE+OOO 2.0300E-002
(* Aluminium dissolution parameters kl padvelO cm/s kc02diss/kaldiss
3.9030E-003 1.3500E+001 1.0000E+OOO
(* cathodic current efficiency
1.0000E+OOO
(* beam speed, (up/down)
2.6000E-001 4.0000E-001
(* tap rate g/s
5.0000E+003
(* composition of ledge ratio, wt% alumina, wt% CaF2
1.4000E+OOO 4.0000E+OOO 4.0000E+OOO
(* density a1203, ledge, cathode, alumina
4.0000E+OOO 2.5000E+OOO 1.6000E+OOO 1.0000E+OOO
(* emmissivity bath, alumina, crust

7.5000E-001 3.0000E-001 5.0000E-001
(* wt% Na 20 AIF 3 NaF in alumina
4.0000E-001 1.0000E+OOO 2.1000E-001
(* Atmospheric pressure N/m2
1.0133E+005
(* % water in alumina, %H in anodes
2.5000E+OOO 9.3000E-002
(* anode stub htc and area
7.5000E-003 3.1400E+002
(* collector bars htc and area
6.5000E-003 2.1550E+003
205
(* htc cathode side and bottom
6.1400E-005 7.0000E-005
(* temperature of cathode layers

9.7271 E+002 9.9825E+002
(* last anode changed
7
(* temperature of cold anode
2.0000E+002
(* number of anode slices
2
(* thicknesses of anode layers
1.0000E+OOO 3.9000E+001
(* factors a, b for fraction of maximum current drawn f=a*exp(bt)
1.7000E-004 6.6900E-005
(* anode #1 f, fraction of current drawn xf thickness of freeze on anode
1.0000E+OOO O.OOOOE+OOO
(* Temperatures in anode zones Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #2f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #3 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #4 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #5 f, xf
(* Tt, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #6 f, xf
(* Tf, T1, T2, TTop
206
9.6571 E+002 9.5426E+002 9.4280E+002 3.9225E+002
(* anode #7 f, xf
1.8224E-003 4.6010E+000
(* Tf, T1, T2, TTop
5.6439E+002 5.4695E+002 5.2978E+002 2.9026E+002
(* anode #8 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #9 f, xf
1.0000E+000 O.OOOOE+OOO
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #10 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #11 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #12 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #13 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #14 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #15 f, x
1.0000E+000 O.OOOOE+OOO f
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #16 f, xf
207
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #17 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* anode #18 f, xf
(* Tf, T1, T2, TTop
9.6574E+002 9.5433E+002 9.4292E+002 3.9230E+002
(* output parameters
(* plot frequency tenth minutes
10
(* tap file name
al6c3.tap
(* data file name
al6c3.dat
(* Data file output
ON
(* Tap file output
ON
(* Printer output
OFF
(* Autotap switch
ON
(* tap frequency hrs
24
(* time to next tap hrs
22
(* AIF3 addition frequency hrs
24
(* time to next AIF3 addition hrs
22
(* Anode Change frequency mins

1920
(* Time since last anode change

595
208
(* Time for anode out of bath mins
4
(* Simulation Time hours
1.7000E+002
(* feedtimer
25
(* target call em
19
(* pad target em
14
A.2 The User Interface for the Simulation Program
Although the initialisation files allow all the parameters in the model to be
initialised, it is a cumbersome method for dealing with those variables that
most commonly require changing, for example feed times or target
voltage. It also makes the program difficult to operate by anyone without a
very detailed knowledge of the initialisation files. In order to facilitate the
use of the software, a front end menu system was developed to allow the
main operating parameters to be easily changed.
The menu system is accessed by function keys from the top level menu
shown below.
Aluminium Reduction Cell Simulation
Control version CELTROL
Model Version SIXC
Fl Output Options
F2 Tapping Schedule
F3 Chemical Additions
F4 Control Variables
FS Model Variables
F6 Run
F7 Quit
209
The version of the control algorithms and the cell model linked into the
program is listed at the top of the screen for reference. The functions keys
allow access to the following menus.
A.2. 1 Output options
The output options menu controls the data that is stored by the program,
the names of the data files used for storage and the length of the
simulation. The frequency at which data is stored is controlled by the Print
Data option which allows the data to be stored as frequently as every
second. This can be used to increase the resolution of simulated events
such as anode effect or to reduce the amount of data stored for longer
runs.
The simulation will also produce a voltage trace with a three minute
interval in the same format as that produced by Anglesey Aluminium
control and datalogging system. This can be directed to a printer or to file.
Output Options
Simulation Time 170.0 hrs
Metal Tap Data to Disc ON
Tap File is alsdata\a16d1.tap
Cell Data to Disc ON
Data File is alsdata\a16d1.dat
Print Data every 1.0 mins
Voltage Trace to Printer OFF
PG UP MAIN SPACE MAIN PG DN TAP
In all these menus, the variable may be changed by pressing the

highlighted letter key. The PgUp and PgDn keys access the previous or
next menu respectively. The Space Bar returns the user to the main
menu.
210
A.2.2 Tapping schedule
The parameters pertaining to the tapping of aluminium from the cell can
be specified from this menu.
Metal Tap Schedule
Metal Tap Rate 5.0 kg / s
Target Pad 14 cm
Automatic Tapping ON
Tap Frequency 24 hrs
Time To Next Tap 20 hrs
Target Call 19 cm
PG UP OUTPUT SPACE MAIN PG DN ADD
The metal tap rate specifies the rate at which the molten aluminium is
removed from the cell into the crucible. The remaining parameters pertain
to the automatic tapping routine. When the routine is enabled the tapping
sequence will be initiated at the specified tap frequency. The amount
removed is determined by the deviation of the pad depth from the target
pad and is based around the target call parameter. This is the amount of
aluminium that should be removed from the cell, expressed as the
required depth in the crucible, if the pad depth is at the target.
211
A.2.3 Chemical additions
The chemical additions menu allows the user to control the amount and
frequency of aluminium fluoride added to the cell to maintain the
composition of the electrolyte. It is intended that this menu could be
extended to include the addition of other compounds, fo r example lithium
fluoride should this be required.
Chemical Additions Menu
Bagged A1F3 Addition Frequen cy 24 hrs
Time To Next A1F3 Ad dition 20 hrs
Mass A1F3 Added 25 . 0 kg
Time Since Last Ano d e Change 19.92 hrs
Anode Change Fr e que n cy 32 hrs
PG UP TAP SPACE MAIN PG DN CNTRL
This menu also allows the frequency of anode change to be specified.
A.2.4 Control variables

The control variables menu IS a function key driven sub menu that
pertains to the control system in the program. This menu is imported from
the control module and the menu items available will be specific to the
version of the control system that is implemented. For example, a control
system incorporating feed back control loops would include menu options
for the controller settings.
Control Variables Menu
Control version CELTROL
Fl Current and Voltage
F2 Cell Type
F3 Main Simulation Menu
In the standard GELTROl version, this menu allows access to the

following sub menus
212
A.2.S Current and voltage
This menu allows the line current and initial target voltage to be specified
for the simulation.
Line Current 151. 0 kA
Target Voltage 4.39 V
PG UP CNTRL SPACE CNTRL PG ON CELL
A.2.6 Cell type
The implementation of the CElTROl system is different for half-break and

point feed cells. The version of the control module linked into the program
therefore defines the cell type that can be simulated. The cell type IS
displayed for reference.
Cell Type
Cell Type is Half break
Mass of Alumina Fed 23.0 kg/dump
Base Feed Time 67 mins
Time to Next Feed 47 mins
Number of Breaks at Feed 2
Number of Dumps at Feed:
Duct End 4
Tap End 4
PG UP VOLTS SPACE CNTRL PG ON MAIN
The basic alumina feed control parameters are also specified from this
menu. The base feed time is the nominal time between feeds but this may
be altered by the control system, for example during search or starve
cycles.
A.2.7 Model variables

Just as the control variables sub menu is imported from the control
module, so the model variables sub menu is imported from the model
213
module. Should features be added to the model that require user input it is
then a simple matter to include additional menu items the menu system
without losing the general menu structure. Only the model module need
then be altered.
Model Variables Menu
Model version SIXC
Fl Bath Composition
F2 Cell Geometry
F3 Main Simulation Menu
In version SIXC of the aluminium reduction cell model, there are two menu
items. These allow the following parameters to be changed.
A.2.B Bath composition
The bath composition menu allows the initial composition of the electrolyte
and the bath temperature to be specified. The alumina is specified in three
states, as a weight percentage of dissolved alumina, as a mass of alumina
held as agglomerates in suspension and as a mass of alumina in sludge
pockets around the edge of the cell.
Initial Bath Composition
Bath Temperature 970.9 C
Wt Ratio NaF / A1F3 1. 23
Wt% Calcium Fluoride 5.1 %
Wt% Alumina in solution 4.01 %
Mass of Alumina in Suspension 0 . 0 kg
Mass of Alumina as SLudge 0.3 kg
PG UP MODEL SPACE MODEL PG DN GEOM
214
A.2.9 Cell geometry
The cell geometry menu allows the initial shape of the frozen sidewalls
containing the electrolyte to be specified. As well as requiring the average
thicknesses of the ledge in the regions of the bath and the metal pad the
depths of the electrolyte and the aluminium pad must be specified.
Auto calculation options have been included to allow the model to be

initialised where the ledge data is not available. This assumes that at the
start of the simulation, the cell is at steady state and a heat balance is
performed to calculate the thickness of the ledges.
The final parameters on this menu are the anode cathode distance and
the operating volts at the start of the simulation. These two parameters
are interdependent and the user can specify which should be calculated
by the program.
Initial Cell Geometry
AUTO CALCULATE OPERATING VOLTS
AUTO CALCULATION OF STEADY STATE LEDGES OFF
Bath Depth 18.8 em
Pad Depth 12.4 em
Ledge Thickness at Bath 8.6 em
Ledge Thickness at Pad 4.6 em
Anode to Cathode Distance 4.8 em
Operating Voltage 4.39
PG UP BATH SPACE MODEL PG DN MAIN
215
Appendix B Output from the Simulation Program
In addition to continuously recording the conditions with in the simulated
cell in data files on the computer disc, the simulation outputs many of the
parameters to the computer screen. This allows the user to follow the
progress of the simulation and to make changes to some of the
parameters should this be required. The format of the output screen is
shown below.
Aluminium Reduction Cell Simula t i on
Mo del ve rsi on SIXC Control v ers ion CELTROL
Day
TIME DATA FILE IS alsda t a \ a1 6e l.dat
SRCH STRV AVC AET TAP
x CELL TYPE Ha lf-Break

x 0 X X
TapFile alsdata \ a16e1. t ap MASS OF ALUMINA FED 23.0 kg / dump
LINE CURRENT 151 . 0 kA

AUTOTAP ON
CALCIUM FLUORIDE 5.11 %
LAST TAP
Wt RATIO NaF / A1F3 1 .23 %
TARGET PAD 14 em
Cel l Conditions a t 0 : 2 : 38
DAY BATH PAD TAPPED
em em kg
OPERATING VOLTS 4 . 39 V
BATH TEMPERATURE 97l C
PAD DEPTH 12 . 4 em
BATH DEPTH 18.8 em
PAD LEDGE 4 . 6 em
BATH LEDGE 8 . 6 em
4 . 8 em
ACD
WT% AL 2 03
3.95 %
PRINTER OF F AVC ON BASE TARGET VOLTAGE 4.39
T Auto T a p V TARGET VOLT S FlO OTHER

ALT - A AVC P PRINTER Q QUIT S SCREEN
" t' th t is generally available to

The screen is designed to provide Informa Ion a
the operators at the cell, either from the GELTROl panel or o~
'r I parameters that It
chalkboards at the front of the cell. Some add Ilona
216
would be desirable to monitor continuously, for example the alumina
content of the electrolyte, are also displayed.
In the top left hand corner of the screen, the status of the various control
routines is continuously updated. Reading from left to right these are
SRCH Search routine

STRV Starve routine
AVC Automatic voltage control
AET Anode Effect Termination
TAP Tapping routine
An X below the mnemonic indicates that the routine is inactive. In the

example above, only the AVC routine is on.
It is possible to change some of the operating conditions interactively,

using the keystrokes indicated on the bottom line. These allow the user to
enable\disable the AVC routine, to manually tap the cell by toggling the
autotap switch, to toggle output to an on-line printer and to change the
target voltage for the cell.
The ability to change operating conditions during the simulation allows the
user to investigate the behaviour of the cell to disturbances in the system.
For example, use of the manual tapping option would allow the effect of
over or under tapping to be investigated.
217
Appendix C. Graphical Presentation of Results
A necessary part of the simulation of a process is the ability to present the
data in an intelligible format. For dynamic simulation this generally
requires the process variables to be displayed graphically against time.
In the simulation of aluminium cells, the process is being investigated over

a long time scale, (typically 5-15 days) and because events lasting only
minutes, eg anode effect, must be logged a large amount of data is
produced. As the simulation takes along time to run it is desirable to run
the simulation on the fastest machine available and to post process the
graphs on other available machines.
Whilst there are many sophisticated spreadsheets and graphics packages

available most are unable to deal with the quantity of data produced by
the aluminium simulation. In order to interpret this data it was therefore
necessary to develop a graphics package capable of manipulating the
data and producing high quality output. As well as being able to deal with
the quantity of data generated, this approach has the advantage of
producing graphs in any desired format and is not limited to the rigid types
of a commercial spreadsheet or graphics package.
The program was written in Modula 2 to maintain the modular nature of

the main program and also to allow the use of text and file modules
previously written. Much of the interpretation of the simulation data
requires the comparison of the trends of process variables at specific
times throughout the simulation. The graphics package was designed to
allow up to three graphs to be plotted on one page with a maximum of four
data sets per graph. In order to investigate particular features of the
model, areas of the graph can be zoomed into and replotted in more
detail.
In order to optimise the speed of graphical analysis the program loads as

much data as possible into memory and continues from disk once this
memory is full. This greatly enhances the speed at which graphs are
drawn as much of the time taken in plotting a graph is due to reading data
from disk.
218
Once the required graphs have been produced and investigated on
screen, hard copy may be obtained on a Hewlett Packard plotter.
219
References
Aarebrot, E. et ai, Metall. vol 32, no 41, 1978
Abramov et ai, "Theoretical Principles of the Electrometallurgy of

Aluminium", Moscow, 1953
Arita, Y., Urata., N. and Ikeuchi, H., "Estimation of Frozen Bath Shape in
an Aluminum Reduction Cell by Computer Simulation", AIME Light Metals,
1978,pp59
Asbj0rnsen, O. A., Andersen, J. A., "Kinetics and Transport Processes in

the Dissolution of Aluminium Oxide in Cryolite Melts", AIME Light Metals
1977, pp137-152
Aylward, G. H. and Findlay T. J. V., SI Chemical Data, 2nd Edition, The

Jacaranda Press, 1974
Azbel, D.S. and Lee, S.L., "Two-Phase Momentum Heat Transfer

Chemistry", Process Eng. Syst., vol 1, p159, 1978
Bagshaw, A. N., Kuschel, G., Taylor, M. P., Tricklebank, S. B., and Welch,
B. J., "Effect of Operating Conditions on the Dissolution of Primary and
Secondary (Reacted) Alumina Powders in Electrolytes", AIME Light
Metals, 1985, pp649-660
Belyaev et ai, Izv, Vyssh. Ucheb. Zav., Tsvetn. Met., vol 4, no 3, 1961,
p67
Bennet and Meyers, "Momentum, Heat and Mass Transfer", McGraw-Hili,

1962
Berge, B., Grjotheim, K., Krohn, C., Naeumann, R. and T0rklep, K., "The
Influence of Operating Parameters on the Current Efficiency in the
Aluminium Reduction Cells", AIME Light Metals, vol 1, 1976, pp23
Bersimenko et ai, Tsvet. Met., vol 46, no 3, 1973, p30
220
Blake, S. R., '~ Model for the Variation of Alumina Concentration in
Industrial Hall-Heroult Cells", MSc Thesis, Department of Chemical
Bratland, D., Grjotheim, K., Krohn, C. and Motzfeldt, K., Journal of Metals,
vol 19, 1967, pp13
Bratland, D., Jose del Campo, J. and Cho, K., "Current Efficiency
Measurements in Laboratory Aluminium Cells. Influence of Bath Volume I
Anode Size and Anode Quality", AIME Light Metals, 1981, pp281-308
Bruggeman, D.A., Ann. Physik, vol 24, 1935, p632
Bullard, G., KACC Research Report, 2nd quarter, 1984
Calandra, A.J., Castellano, C.E., Ferro, C.M. and Cabo, 0., "Experimental
and Theorectical Analysis of the Anode Effect in Industrial Cells", AIME
Light Metals, 1982, pp345-358
Choudhary, G., "Electrical Conductivity for Aluminum Cell Electrolyte

between 9SCf>-102SOC by Regression Equation", J. Electrochem. Soc., vol
120,1973, pp381-383
Dewing, E.W., "Liquidus Curves for Aluminum Cell Electrolyte v.

Representation by Regression Equations", J. Electrochem. Soc., 117,
1970, pp780-781
Dow, D.W. and Goodnow, W.H., "Influence of Operating Variables on

Reduction Cell Bath Temperature", AIME Light Metals, 1972, pp3-20
Ek, A. and Fladmark, G. E., "Simulation of Thermal, Electric and Chemical

Behaviour of an Aluminum Cell on a Digital Computer", AIME Light
Metals, 1973, pp85
Entner, P, Scmidt-Hatting, W., Mitrovic, Z., Gruber, U., Rufer, D. and

221
Evans, J.W., Zundelevich, Y. and Sharma, D., Met. Trans, vol 128, 1981
Fellner, P., Grjotheim, K. and Kvande, K., "Model Calculations of Cryolite

Melts", AIME Light Metals, 1984, pp805
Filimonov, S. S., Kryukova, M. G. and Teplov, S. V., Thermal Physics of

High Temperatures, vol 2, no 6, 1964, pp901
Firsanova et ai, Izx. Vyssh. Ucheb. Zav., Tsvetn. Met., vol 5, no 3, 1962,
p52
Frovlova, E.B., Dobrokhotov, V.B. and Tsyplakov, A.M., Trudy V.A.M.I, 89,
vol 36, 1971
Gerlach, J., Hennig, U. and Kern, K., "The Dissolution of Aluminium Oxide
in Cryolite Melts", AIME Light Metals, vol 1, 1974, pp49-61
Gjerstad et ai, Symposium on Industrial Electrode Processes with Gas

Grjotheim K. et ai, Can. Met. Quart., vol 11, 1972, p295
Grjotheim, K., Haupin, W. E. and Welch, B. J., "Current Efficiency--

Relating Fundamental Studies to Practice", AIME Light Metals, 1985
pp679-694
Grjotheim, K., Kvande, H. and Motzfeldt, K., "Vapor-Liquid Equilibria in the

System NaF-AIF3-AI203'" AIME Light Metals, 1975, pp125-137
Hashimoto, T. and Ikeuchi, H., "Computer Simulation of Dynamic

pp273
Haupin, W.E., '~ Scanning Reference Electrode for Voltage Contours in

Aluminum Smelting Cells", Journal of Metals, Oct., 1971
Haupin, W. E., "Calculating Thickness of Containing Walls Frozen from

222
Haupin, W.E., "Overvoltages Determined by Nonlinear Regression
Analysis of Cell Volts vs. Current Density", AIME Light Metals, 1980,
pp183-190
Haupin, W. E., "Mathematical Model of Fluoride Evolution from Hall-

Trondheim, 1985
Haupin, W. E., "The 4th Int. Course on Process Metallurgy of Aluminium",

Trondheim, 1985
Haupin, W.E. and Frank, W.B., "Comprehensive Treatise of

Electrochemistry", eds. Bockris, Conway, Yeager and White, Plenum
Publishing Corp., New York, 1981, p308
Hertzberg, T., T0rklep, K. and 0ye, H.A., "Viscosity of Molten NaF- AIF3-
AI2 0 3 -CaF2 Mixtures", AIME Light Metals, 1980, pp159-170
Hove, S.J. and Kvande, H., "Centre-Break Alumina Feeding and Sludge
Control of Prebaked Cells", AIME Light Metals, 1982, pp513
Jain, R. K. et ai, "Interaction of Aluminas with Aluminium Smelting

Electrolytes", AIME Light Metals, 1983, pp609
JANAF Thermochemical Tables, NRDS, National Bureau of Standards 37
Johnston, A. R., ''Alumina Crusting and Dissolution in Molten Electrolyte",

Johnston, T.J. and Richards, N.E., "Laboratory and Field Studies with an
Improved Direct Reading Alumina Concentration Meter", AIME Light
Kachanovskaya, I. S., Osivik, V. I. and Kukhotkina, T. N., "Phase

Transformations and Alumina Dissolution in Cryolite", Tsvetnye Metally,
No 12, 1971, pp10-23
223
Keller, R., "Alumina Dissolution and Sludge Formation", AIME Light
Metals, 1984, pp513
Kent, J. H., "A Review of the Hal/-Heroult Process", Aluminium and

Energy, A 100 Year Partnership Conference and Exhibition, Bangor, Sept
18-19, 1986
Kvande, H. and Rorvik, H., "The Influence of Bath Density in Aluminium

Electrolysis", AIME Light Metals, 1985 pp671-678
Less, L. N., "The Crusting Behaviour of Smelter Aluminas", AIME Light

Metals, 1976, pp315
Levich, V.G, "Physicochemical Hydrodynamics", Prentice Hall, New

Jersey, 1962, pp690-698
Lewis, R. A., "Technical Fundamentals of the Aluminum Reduction Cell

Lillebuen, B., and Mellerud, Th., "Current Efficiency and Alumina

Lillebuen, B., Ytterdahl, S. A., Huglen, R., and Paulsen, K. A., "Current
Efficiency and Back Reaction in Aluminium Electrolysis", Electrochimica
Acta, vol 25, No, 2, February 1980, p131
Maeda, H., Matsui, S., and Era, A., "Measurement of Dissolution Rate of
Alumina in Cryolite Melt", AIME Light Metals, 1985 pp763-780
Manaktala, S., "Process Control Techniques for Reduction Cell

Operation", AIME Light Metals, 1971, pp165
Matiasovsky, K., Danek, V. and Malinovsky, M., "Effect of LiF and Li~/F6
on the Electrical Conductivity of Cryolite-Alumina Melts", J. Electrochem.
Soc., 116, 1969, pp1381-1383
224
Newitt, Dombrowski and Knellman, Trans. Inst. Chem. Engr., vol 32, 1954
p244 '
Paulsen, K.A., Huglen, R., Andersen, J.A., Lillebuen, B., and Ytterdahl ,
S.A., "Variation of Side Lining Temperature, Anode Position and
CurrenWoltage Load in Aluminium Reduction Cells", AIME Light Metals,
1980, pp325-339
Peacey, J. G. and Medlin, G. W., "Cell Sidewall Studies at Noranda

Plimley, R. E, "Process Modelling and Optimisation", Electrochemical

Process Engineering Course, University of Newcastle upon Tyne, 1984
Poole, R.T., Etheridge, C., ''Aluminium Reduction Cell Variables and

Operations in Relation to Current Efficiency", AIME Light Metals, 1975,
pp163-183
Qiu, Z. and Fan, L., "The Rate Determining Step of Metal Loss in Cryolite-
Alumina Melts", AIME Light Metals, 1984, pp789
Reynolds Metals Company, "Alumina Concentration Meter", U.S. Patent

3,471,390, October 7, 1969
Richard, M. C., "Use of a Mathematical Cell Model to Determine Cell

Parameter Design Changes for Production Maximization", AIME Light
Metals, 1975, pp95
Robl, R. F., Haupin, W. E., and Sharma, D., "Estimation of Current

Rolseth, S. and Thonstad, J., "On the Mechanism of the Reoxidation

Reaction in Aluminum Electrolysis", AIME Light Metals, 1981, pp289
Saksvikronnig, T., Valsvik, G., and Hove, S. J., "Anode Effect Quenching
by Anode Tilting", AIME Light Metals, 1982, p553-558
225
Schmitt, H., "Extractive Metallurgy of Aluminium", vol2 Intersc' P b N
' I. U., ew
York, 1963, p169
Sides, P.J. and Tobias, C.W., J. Electrochem. Soc., vol 129, 1982, p2715
Sidorov, L.N., Kolosov, E.N. and Shol'ts, V.B., Russian Journal of Physical
Chemistry, 42, pp1382-1386, 1968
Solheim, A. and Thonstad, J., "Heat Transfer Coefficients between Bath

and Side Ledge in Aluminum Cells. Model Experiments", AIME Light
Metals, 1983, pp425
Sulmont, B. and Hudault, G., "Application of a Thermoelectric Model to the

Investigation of Reduction Cell Thermal Equilibrium", AIME Light Metals,
1978,pp73
Szeker, Acta Tech. Acad. Sci. Hung., vol 10, 1955, p91
Tabereaux, A.T. and Richards, N.E., "An Improved Alumina Concentration

Meter", AIME Light Metals, 1983, p495-506
Thonstad, J., "The 4th Int. Course on Process Metallurgy of Aluminium",

Trondheim, 1985
Thonstad, J. et ai, Proceedings, 3rd Czech. Aluminium Symposium,

Banska Bystrica, 1976, p88
Thonstad, J., Johansen, P., and Kristensen, E. W., "Some Properties of

Alumina Sludge", AIME Light Metals, 1980, pp227-241
Thonstad, J., Nordmo, F. and Paulsen, J. B., "Dissolution of Alumina in

Molten Cryolite", Metallurgical Transactions, vol 3, February 1972, pp403-
407
Thonstad, J. and Rolseth, S., Proceedings of 3rd ICSOBA Conference,

Nice, 1973, p657
226
:honst~d,. J. and ROlset~, S., "Equilibrium between Bath and Side Ledge
In AlumInIum Cel/s. BasIc Principles", AIME Light Metals, 1983, pp415
Thonstad, J., R0nning, S. and Entner, P., "Formation of Bottom Crusts in

Aluminium Pots. A Laboratory Study", AIME Light Metals, 1982, pp485
Vasiliadis, C. G., "Mass Transfer Studies related to Aluminium

Electrolysis", MSc Dissertation, Oct 1981, University of Newcastle upon
Tyne
Vetyukov, M. M. et ai, Physical Chem. and Electrochem. of Molten Salts

and Slags, Kiev, 1969, pp367
Vetyukov M. M. et ai, Isv. Vyssh. Uchebn. Zaved., Tsvetn. Metall., 1978,

p62
Waddington, J. and Pearson, T. G., "Electrode Reactions in the Aluminium

Reduction Cel/", Discussions, Faraday Society, 1947, vol 1, pp307-320
Welch, B.J., "Energy Balance-Transient Thermal Effects During Cell

Operation", The 4th Int. Course on Process Metallurgy of Aluminium,
Trondheim, 1985
Wilson, C.A and Tabereaux, A.T., "Alumina Control in Center-Break

Cel/s", AIME Light Metals, 1983, pp479-493
Winkhaus, G., "On the Dissolution Rate of Alumina in Cryolite Melts", 99th
AIME Annual Meeting, Denver,1970, Paper A70-25
Wittner, H., "Working Methods for High-Current Efficiency and Flexible

Amperage", AIME Light Metals, 1975, pp139-150
Wittner, H. and Giulini, Gebr., "How to run a Smelter with Current

Efficiency higher than 90%-Scientific and Practical Basis", AIME Light
Metals, 1977, pp153-161
Tabereaux, A.T. and Richards, N.E., '~n Improved Alumina Concentration

Meter", AIME Light Metals, 1983, pp495-506
227
Zuca, S., Herdlicka, C. and Terzi, M., "On Porosity-Overvoltage Correlation
for Carbon Anodes in Cryolite-Alumina Melts", Electrochimica Acta, vol 25,
February 1980, pp211-216
228
Nomenclature
aJ activity of species j in the electrolyte
A area
ACO anode to cathode distance
Cj concentration of species j in the electrolyte
c~ saturation concentration of species j
J
%CE cu rrent efficiency as a percentage

Cf friction factor
Cp specific heat capacity
CR Cryolite ratio = mole ratio NaF/AIF3
Cs factor for skewness of velocity profile
d inter electrode distance
de critical dimension for Reynolds number in the bath
O·J diffusivity of species j in the electrolyte
Oem equivalent convective diffusivity for metal species in bath
Om mean molecular diffusivity of metal species in bath
E electrical potential
E0 thermodynamic equilibrium potential for a reaction
F Faraday's constant
fe fraction of anode surface covered with bubbles
FE mass of fluoride entrained in the pot gases per tonne of
aluminium
mass of gaseous fluoride produced per tonne of
aluminium
mass of particulate fluoride in the pot gases per tonne of
aluminium
g acceleration due to gravity
gap gap between banks of anodes
GE scrubber efficiency for gaseous species
Gr Grashof number
~G Gibbs free energy
~Hfus latent heat of fusion
~HR',J heat of reaction j
h heat transfer coefficient
Hdiss , j heat of dissolution of species j
HE hooding efficiency
H!J heat of formation of compound j
Hm depth of metal pad
229
total current
I·J current associated with reaction j
current density
l'
J
partial current density associated with reaction j
K electrical conductivity
k rate constant
k·J thermal conductivity of component j
Kd equilibrium constant for dimerisation of NaAIF4 vapour
kL mass transfer coefficient
Kp constant of proportionality
kt turbulent kinetic energy
L physical dimension for description of system
M·J mass of component j
n number of anodes
N stirrer speed
Vf velocity of anode gases
Nu Nusselt number
Pj partial pressure of species j
Pb barometric pressure
PE scrubber efficiency for particulates
Pr Prantl number
Pv total vapour pressure
q heat flow
Ok volumetric flowrate of stream k
R·I rate of reaction i
R·J electrical resistance of component j
R universal gas constant
Ra Rayleigh number
rAI 20 3 radius of AI 20 3 agglomerate
rate of convective transport of aluminium into the
rAlconv
electrolyte
aluminium dissolution rate
rAldiss
Rb weight ratio NaF/AIF3
Re Reynolds number
Rm radius of rotation of metal pad

rate of convective transport of metal into the electrolyte
rm
theoretical maximum rate of production of aluminium
rt
Sc Schmidt number
230
SCt turbulent Schmidt number
Sf fraction of feed becoming sludge
Sh Sherwood number
T temperature
To reference temperature
ta thickness of bubble layer at anode
td dead time
tf fixed feed time
tremelt time taken for electrolyte frozen onto a surface to melt
ts activation time for solenoids
U overall heat transfer coefficient
Ut turbulent viscosity
V volume
W·J molecular weight of species j
wt weight percent alumina in electrolyte
x thickness of a layer or zone
Xj mole fraction of species j
xa anode width
XA1F3 AIF3 in electrolyte expressed as a weight percentage
excess of that in pure cryolite
anode length
distance normal to the interface
Charge number
Tafel constant
Tafel constant
bubble diameter
gas fraction in free bubble layer
emissivity in centre trench
11 overpotential
f inter electrode gap
A'J latent heat of fusion of species j
Il viscosity
v interfacial velocity
e surface coverage by bubbles at the anode
p density
Stefan Boltzmann constant
molar volume of species j
231
suffixes
A Anode
ac alumina cover
af frozen layer at anode surface
AI Aluminium
AI 20 3 Alumina
A13+ Aluminium ions
AIF3 Aluminium fluoride
As anode surface
b bath
ba bath
bf bath / frozen ledge interface
bt bath / alumina interface
C cathode
cav cavity beneath shields
CO2 carbon dioxide bubbles
conv convective
cov alumina cover on top of anode
ct centre trench
d dimer Na2AI2 Fa
diss dissolution
e electrolyte
ex exposed bath
ext external
F fluoride ions
f freeze
g gas
h horizontal component
H2O water
HF Hydrogen fluoride gas
hole hole in the crust
I liquid
m monomer NaAIF4
mf metal pad / frozen ledge interface
NaAIF4 Sodium Aluminium Fluoride
0 oxide reaction or ions
sludge alumina as sludge at the cathode
stub anode stub
232
suspn alumina in suspension in the electrolyte
tc centre trench / cavity interface
v vertical component
vol volatilisation
W water
sat saturation
233

Wright 93

Uploaded by

Copyright:

Available Formats

Wright 93

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

Wright 93

Uploaded by

Copyright:

Available Formats

The Dynamic Simulation and Control of Aluminium

Department of Chemical and Process Engineering

Submitted in partial fulfilment of the requirements for the degree of Doctor

NEWCASTLE UNIVERSITY LIBRARY

The Hall-Heroult process for the electrolytic production of aluminium from

The introduction of computer control and data logging systems has

In this work, a mathematical model for the dynamic simulation of

The model is evaluated against data from Anglesey Aluminium's half-

My supervisor, Dr Allen Wright for his continual support, encouragement

The UK Science and Engineering Research Council and Anglesey

2.0 Process Modelling 15

2.1 Model Development 16

2.2 Types of Model 18

2.3 Choice of Model 20

3.0 Mathematical Models of Aluminium Reduction Cells 21

3.1 Static Models 21

3.2 Dynamic Models 22

3.3 Models for the Dissolution of Alumina 23

4.0 Development of a Dynamic Model 75

4.10 Heat Losses 124

4.11 Pot Heat Balance 132

4.12 Frozen Sidewalls 136

4.13 Anode Change 139

4.14 Phyico-Chemical Relationships 143

4.15 Summary 145

5.0 The Control System 149

5.1 Alumina Concentration Control 150

5.2 Removal of Aluminium Product 150

5.3 Electrolyte Composition 151

5.4 Anode to Cathode Distance 151

5.5 Current Distribution 152

5.6 Voltage and Temperature Control 152

5.7 Manual Control 153

5.8 Manual Control of Alumina Concentration 154

5.9 Manual Control of Voltage 154

5.16 Alumina Concentration Control 162

5.17 Anode Effect Termination 165

5.18 Point Feeder Control Strategies 165

5.19 Anode Effect Termination Routine 166

5.20 Alternative Strategies 166

5.21 Alumina Concentration Control 166

5.22 Open loop Alumina Feed Strategies 167

5.23 Closed loop Alumina Feed Strategies 168

6.0 Dynamic Simulation of an Aluminium Reduction Cell 173

6.2 Initialisation 174

6.3 Presentation of Results 175

6.4 Numerical Solution of Differential Equations 176

6.5 Simulation of Current Control System (CElTROl) 176

6.6 Validation of the Simulation 177

6.7 Plant Trials 190

6.8 Conclusions 191

7.0 Recommendations For Further Work 193

7.1 Inferential Control 193

7.2 Development of the Real Time Simulation (RTS) 193

7.3 Tuning the Model for Control 198

A Initialisation of the Simulation Program 203

A.2 The User Interface for the Simulation Program 209

B Output from the Simulation Program 216

C Graphical Presentation of Results 218