Aluminium Electrolysis
Ingo Eick
1, Jinsong Hua
2, Kristian E. Einarsrud
3, Peter Witt
4, Wei Bai
51 Hydro Aluminium, 41468 Neuss, Germany
2 Institute for Energy Technology (IFE), 2027 Kjeller, Norway
3 Sør Trøndelag University College (HiST), 7004 Trondheim, Norway 4 CSIRO Minerals Resources Flagship, 3169 Melbourne, Australia 5 SINTEF Materials and Chemistry, 7465 Trondheim, Norway
• Hydro Aluminium metal production (03-06)
• Aluminium electrolysis process (07-14)
• Modelling assumptions (15-18)
• Steady state metal pad profile and MHD prediction (19-27)
• Transient bubble and chemical reaction flow model (28-38)
• Steady state full cell bath flow model (39-44)
• Transient Alumina reaction and distribution model (45-50)
All production figures are from 2012
Bauxite/alumina Smelters
Remelters
North America
•114 000 mt smelting capacity in 1 smelter
•369 000 mt remelt capacity in 6 remelters
Australia:
•228 000 mt primary smelting capacity in 2 smelters
Carribean / South America:
•1 890 000 mt alumina production in 2 alumina refineries
•2 700 000 mt bauxite production in 2 bauxite mines
•1 MoU new alumina JV project (Hydro share: 20%)
Europe:
•1 457 000 mt primary smelting capacity in 7 smelters
•359 000 mt remelt capacity in 6 remelters
Middle East:
• 300 000 mt smelting capacity in 1 smelter
Location Share Start-up Technology Layout amperage Cells Operating amperage Increase Ardal AI 100% 1970 HAL150 160 kA 216 209 kA 131% Ardal AII 100% 1978 HAL170 172 kA 110 221 kA 128% Hoyanger L2 100% 1981 HAL230 230 kA 80 282 kA 123% Husnes L1/2 100% 1965 EPT10 140 kA 400 150 kA 107% Karmoy L3 100% 1987 AP18 175 kA 288 232 kA 133% Sunndalsora SU3 100% 1968 HAL150 175 kA 184 203 kA 116% Sunndalsora SU4 100% 2003 HAL275 275 kA 340 314 kA 114% Rheinwerk L1 100% 1962 CA120 120 kA 162 180 kA 150% Rheinwerk L2 100% 1965 CA125 125 kA 156 180 kA 144% Rheinwerk L3 100% 1965 CA125 125 kA 156 180 kA 144% Qatar L1/2 50% 2010 HAL300 300 kA 704 355 kA 118% Ziar Nad Hronom L3 50% 1996 HAL230 230 kA 226 255 kA 111% Sept-Iles L1 20% 1992 AP30 295 kA 288 365 kA 124% Sept-Iles L2 20% 2005 AP35 335 kA 312 365 kA 109% Tomago L1 12% 1983 AP18 180 kA 280 225 kA 125% Tomago L2 12% 1993 AP18 180 kA 280 225 kA 125% Tomago L3 12% 1998 AP20 197 kA 280 225 kA 114%
Operational excellence
Capacity creep New design
Simulation
, a main tool, supporting the continuous improvement process
• Reduce variations • Reduce emissions • Reduce energy consumption • Step change • New design features• Very low energy consumption
• Neutral CO2 foot print
• Increase production on same technology base
Main principle:
cell 1
cell 2 ...
Al2O3 powder Al2O3 powder
Cryolite melt (bath) (960°C) CO2 (g) CO2 (g) Current 150 – 600 kA Carbon anode Al (l) Cathode Cryolite melt (960°C) Carbon anode Al (l) Cathode
Magnetic effects:
A series of 150 – 350 cells are connected to current cycle in a potroom 150 to 600 kA applied to the conducting system (busbar) of the potline Strong magnetic field of busbar
Magnetisation of steel shell
Lorentz forces on liquid metal and bath Induced fields
Induced current
Time scale: < 1 milli second
Length scales: 1 km long potroom, 15 mm thick steel shell
Magneto-Hydrodynamics (MHD): 200 150 100 50 0 cm 1:20 1:10 1:5 1:10 50 40 30 25 20 10 0 cm 1) Generates a manifold overlaying magnetic field partly shielded by the shell
2) Resulting in Lorenz forces, stirring the metal and electrolyte and piling up a metal pad profile
A complex current path across different material designed to reduce magnetic effects
Main reaction: reduction of alumina based on 6 species (Al2O3(sol), Na2Al2O2F4, Na2Al2OF6, AlF3, NaF, Na)
Alumina feeding process Alumina dissolution
Equilibrium reaction
Anode boundary layer reaction Cathode boundary layer reaction
Local species concentration will affect properties of bath like viscosity, resistivity, ... Time scale: Built-up concentration layer at anode / cathode < milli seconds,
Feeding cycles ≈ hours
Length scale: Concentration layer < 1 mm, species transport = cell length
Aluminium Electrolysis Process
Al2O3(sol)+ 3 NaF + AlF3 →Na2Al2O2F4 Na2Al2O2F42-+ 2 NaF + 2 AlF
3 ↔ 2 Na2Al2OF6
2-2 Na2Al2OF62-+ C →4e- + 4 AlF
3 + 4 Na + CO2
Bubble flow:
Bubble nucleation (not fully understood)
Bubble growth
Bubble coalescence
Bubble transport
Bubble releaseReleasing CO2 with a strong stirring effect
Time scale: Release frequency ~ 2-5 seconds
Length scale: nucleon ≈ 0.5 mm, rising plug > 10cm
Metal layer
Carbon anode Bath
Bath flow: based on bubble draft and MHD
Metal surface profile
Metal surface movement
Bubble draft
Lorentz forces around bubble
Convection flow an ledge sides
Bath film around metal (not fully understood)Time scale: bath speed ~ 10 – 20 cm/s
Length scale: cell length ~ 10 - 20 m, height of reaction zone < 40mm Aluminium Electrolysis Process
Time and length scales
•
Magnetic field and species reaction quite fast separable•
Bath and bubble flow in same domain how to interlink•
Species transport with longer time scale separable 10 decades in time1)
Magneto-hydrodynamics (MHD)•
Magnetic field is establishing fast•
Overall current distribution in cell long term stable (anode cycle ≈ 24 days)•
Metal pad is mainly defined by overall current distribution•
Metal flow speed is mainly defined by overall current distribution But•
bubble movement => local current distribution below anode•
MHD instabilities can initiate surface waves on metal pad Steady state approach for magnetic field, metal pad profile and speed. Output forwarded to next step.
2) Chemical reaction of alumina reduction
•
Fast reactions and large diffusion coefficients Concentration layers at anode and cathode can be solved as boundary equations.
But
•
Transient feeding pattern (underfeeding / overfeeding) required for cell controll•
Distribution of species concentration by advection across 20 m long cell3) Bubble flow
•
Transient approach on bubble formation, coalescence and draft required•
Detailed anode geometry of rounding and slots are relevant•
Bubble draft forces in bath stronger than advection from metal surface Flow pattern at single anode, based on overall stable global metal pad and magnetic field, can be simulated separately and transferred as volume draft forces and turbulent viscosity into a full cell model
60% 65% 70% 75% 80% 85% 90% 95% 100% 105% 110% 0 4 8 12 16 20 24 Current pick-up
Operation days of anode 4) Full cell bath flow
•
Due to stationary MHD and bubble draft steady state approach sufficient•
Detailed geometry of ledge profile requiredprofile and MHD
prediction
Modelling approach:
Flow field
• Two phase flow model approach for bath layer and metal pad
• The volume of fluid (VOF) method is used to simulate the dynamics of the immiscible two-fluid system
Magnetic field:
• Electric current density J is calculated
• Electrical potential distribution is given as current conservation
• Induced magnetic field (Bind) is obtained by solving magnetic potential vector (A)
• The Lorentz force calculated Turbulence model:
• Standard k-ε turbulence model is solved for the turbulent viscosity that is included in the effective viscosity of fluids.
Modelling approach:
Flow field
• Two phase flow model approach for bath layer and metal pad
• The volume of fluid (VOF) method is used to simulate the dynamics of the immiscible two-fluid system
where , and FE is the Lorentz force.
• The secondary phase volume fraction is accomplished by solving, the continuity equation, and the primary phase volume fraction is obtained by the continuity constraint
Steady state metal pad profile and MHD prediction
Magnetic field:
• Electric current density J is calculated as
where = electrical potential, B = magnetic flux density and = electrical conductivity
• Electrical potential distribution is given as current conservation
• Induced magnetic field (Bind) is obtained by solving magnetic potential vector (A)
• Induced magnetic field:
Turbulence model:
• Standard k-ε turbulence model is solved for the turbulent viscosity that is included in the effective viscosity of fluids.
Programming structure:
Implementation in ANSYS Fluent by
• User defined scalar for: , AX, AY, AZ
• User defined functions for: B, J, FE
• User defined memory for variable storage
Steady state metal pad profile and MHD prediction
Implementation
Mesh adjustments:
• Dynamic tracking of Bath/Metal interface using Fluent VOF (volume fraction 0.5) and sliding mesh approach to adjust anode bottom shape to metal pad profile
Boundary conditions:
• Top surface of bath with wall function
• Realistic side and end ledge profile with standard wall function
• Rigid surface between metal and bath assumed
• Mesh: 35 mm wide gaps between anodes resolved for 4 x 12 m cell geometry
Steady state metal pad profile and MHD prediction
Simulation result:
Metal pad profile (interface between bath and metal layer)
• Introduces an inclination of anode bottom due current dependent anode consumption
• Metal pad profile transferred to full cell bath flow model
Velocity field:
Metal surface speed
• Metal pad surface speed transferred to full cell bath flow model
Steady state metal pad profile and MHD prediction
Result verification:
• Generally good agreement
• Realistic geometry significant
• Induced current significant
• Induced fields time consuming and less significant
• Measurement approach with steel rods
− Rod dissolution profile flow direction
− Dissolved mass / time flow speed
− Very rough integral method
chemical reaction flow
model
Cooperation with
SINTEF, HIST
Modelling approach
This modelling multi-scale and multi-field approach, aiming to fully resolve the behavior of macroscopic bubbles. Unresolved fields are treated by
applicable sub-grid models
In the model the different physical phenomena are considered:
1) Current Magnetic field, Lorentz forces
CO2 generation, Faraday law
2) Species concentrations, especially CO2
3) Phase fractions, PBM
4) Realized bubbles with VOF method
5) Flow field
Microscopic properties (conductivity, surface tension and contact angles) are dependent upon dissolved, dispersed and continuous fields
Modelling approach
This modelling multi-scale and multi-field approach, aiming to fully resolve the behavior of macroscopic bubbles. Unresolved fields are treated by
applicable sub-grid models
In the model the different physical phenomena are considered:
1) Current Magnetic field, Lorentz forces
CO2 generation, Faraday law
2) Species concentrations
3) Phase fractions
4) Realized bubbles with VOF method
5) Flow field
Only metal pad profile is given as geometry profile and velocity boundary condition
Transient bubble and chemical reaction flow model
1) Electro magnetic fields
•
The current density is obtained by solving a Laplace equation for the electrical potential0
where is the electrical conductivity, both phases (gas, bath) and depending upon local species concentrations (Bath = f(ci)).
•
Lorentz forces FE = J x Bo are obtained by the current density J = - . Induced fields and currents are not considered.•
CO2 gas generation is defined to the current density as of Faraday’s equations.2) Dissolved fields
•
Seven species are considered; Al2O3, NaF, AlF3, Na2Al2O2F6, Na2Al2OF4, Na+ and CO2.
•
The behavior of all species is governed by a generic advection-diffusion equation:⋅
on a mass fraction basis, where is the effective
(turbulent) diffusivity and and are source and sink terms due to consumption and production of the th specie.
• The saturation of the bath with dissolved CO2 enables
Transient bubble and chemical reaction flow model
3) Dispersed fields
•
The dispersed field is concerned with small scale bubbles, typically ranging from a diameter of 0.4 mm (nuclei) and up to sizes dictated by the numerical resolution.•
Dispersed bubbles are modelled by means of a discrete population balance model (PBM), which describes the evolution of number densities under coalescence and mass transfer.•
Ten classes were used with a ghost class for transition into the two phase model.Typical 2-5 mm
4 + 5) Continuous fields toward flow fields
•
Continuous (resolved) fields are treated by means of the Volume of Fluid (VOF) method, allowing fordirect simulation of the complex bubble topology present on the anode surface.
•
The VOF-approach is in the current formulation extended in order to allow for dynamic contact angles.•
Turbulence is modelled by the realizable -model. Transient bubble and chemical reaction flow model•
Large scale properties, i.e. density and viscosity are dependent upon the distribution of continuous fields.•
Microscopic properties (conductivity, surface tension and contact angles) are dependent upon dissolved, dispersed and continuous fields.•
Only restricted information and knowledge onconcentration dependent microscopic properties are available.
Implementation
ANSYS Fluent used as simulation software
•
UDF calculating− Lorentz forces
− Species concentration based on reactions
− Material properties based on Species concentration
•
Random bubble nucleation points feed by CO2 of oversaturated bath•
PBM with 10 classes plus ghost class to reflect sub-grid bubbles•
VOF bubble reconstruction with enhanced contact angle and surface tension•
Anode rounding and metal pad profile is given as geometry profile, while velocity from metal interface as boundary conditionSimulation results
Typical simulation on 30 seconds of anode bubble flow with variation of:
•
Current density (CD = 8000/11000 A/m2)•
Anode inclination ( = 2° / 4° )•
Anode cathode distanceCD = 8000 A/m2 CD = 11000 A/m2
2°
4° Gas phase
distribution after 30 sec of flow time Investigation of process parameter
outside today operation conditions possible...
... but huge CPU time of 300 h*) with fix
time steps of 0.001s per simulation needed.
Result verification
Measurement techniques for flow pattern and species concentration are missing in the harsh environment of industrial cells.
Laboratory cell experiment with “large” 10x10 cm anode carried out
•
90 measurement in lab cell with different current density, anode tilting and ACD carried out.•
Recorded bubble release and cell voltage used for comparisonTransient bubble and chemical reaction flow model
Lab cell experiment:
•
Six simulations have been performed with constant surface tension and contactangles, with conditions corresponding to a sub-set of the experiments
Lab cell experiment:
•
Voltage signal with similar behavior e.g. at CD=0.8 A/m2, the predicted frequency is0.54 Hz with amplitude 97 mV, while
corresponding numbers from experiments are 0.44 and 115 mV
•
Overall comparison good•
Simulations show overall higher release frequencies (bubble speed) and lower amplitudes (smaller bubbles), indicating suppression of certain surface phenomenaTransient bubble and chemical reaction flow model
Modelling approach:
Bath Flow Model – Steady State
• Eulerian-Eulerian or two-fluid model
• Conservation equations for phase mass and phase momentum.
• MHD forces included
• Modified κ- turbulence model in liquid phase only.
• Bubble draft and phase turbulence from zero equation model.
• Time averaged gas distributions, gas & liquid velocities and turbulence quantities.
• Chemical reaction model with 6 species applied
Steady state full cell bath flow model
see P. Witt et al, 10thInt Conf on CFD in Oil & Gas,
Modelling implementation:
Realistic geometry required:
• Anodes of different age considered
• Ledge profile of sides and ends
• Metal pad profile
Anode with deep, flat and no slots Alumina Feed Area Free Surface Treated as a degassing boundary Symmetry Plane Anode Base Gas inlet through
Simulation results:
• Velocity field stable against temperature changes
• Velocity field stable against viscosity changes
• Turbulent viscosity 1000 time higher than bath viscosity
• High cross flow speed in area with no slots
Steady state full cell bath flow model
see P. Witt et al, 10thInt Conf on CFD in Oil & Gas,
Simulation results:
• Gas accumulation below anode and in slot visible
• High cross flow speed in area with no slots
• Simulation indicating performance deficit of anode toward end of anode cycle
Results verification:
Extensive work was done in the past with water models but not all effects (MHD, contact behavior, surface tension) could be represented.
Actually
• Only point wise measurement of flow speed possible with steel rod method
• No direct measurement of gas accumulation in cryolite bath available
• No tomographic approach for measuring flow fields available at 960°C
More effort needed on measurement technique
Steady state full cell bath flow model
see P. Witt et al, 10thInt Conf on CFD in Oil & Gas,
model
Modelling approach:
Transient transport model
• Time averaged fluxes used to transport and reaction of all species
• Steady state bath flow field is fixed boundary condition.
• A time marching solution gives spatial and temporal variations in alumina and other species concentration and reaction rates
Transient alumina reaction and distribution model
see P. Witt et al, 10thInt Conf on CFD in Oil & Gas,
Modelling implementation:
Based on bath flow model:
• Geometry
• Mesh
• Flow field as boundary condition
• Alumina feed: 0.25 kg every 80 seconds at 5 spots
• Explicit alumina feeding curve can be applied at every feeding spots
Initial mass
fraction of species needs several iteration to stabiles
Simulation results:
1) High undissolved alumina in feeding area
2) Well distribution of dissolved alumina
3) NaF concentration indicating good performance
Transient alumina reaction and distribution model
see P. Witt et al, 10thInt Conf on CFD in Oil & Gas,
Metallurgical and Process Industries, Trondheim, Norway
1
2 3
Simulation results:
• Undissolved alumina is rapidly distributed in the bath around the anode
• 60 seconds after feeding no fluctuation visible
• Dissolved alumina for anode reaction is stable during feeding cycle 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 20000 20010 20020 20030 20040 20050 20060 20070 20080 20090 20100 Mas s Fr ac ti o n Time [s]
Alumina Point 1 Point 2 Point 3 Point 4 Point 5 Point 6 Point 7 Point 8 Point 9 0.0862 0.0863 0.0864 0.0865 0.0866 0.0867 0.0868 20000 20010 20020 20030 20040 20050 20060 20070 20080 20090 20100 Mas s Fr ac ti o n Time [s]
Na2Al2OF6 Point 1 Point 2 Point 3 Point 4 Point 5 Point 6 Point 7 Point 8 Point 9
Results verification:
• No direct measurement technique of species in cryolite bath available
• No tomographic approach for measuring species concentration available at 960°C
Indirect approach by measuring the changing electrical
resistivity at each anode is a possible option for verification
Transient alumina reaction and distribution model
see P. Witt et al, 10thInt Conf on CFD in Oil & Gas,
For simulation of multi-scale, multi-physic process of aluminium electrolysis it was required to:
• Separating effects and apply as initial condition • Reducing complexity and apply as boundary conditions • Couple models by applying volume sources or
Acknowledgement
This work was supported by RCN project ES497436