Hydrometallurgy 147–148 (2014) 41–53

Contents lists available at ScienceDirect

Hydrometallurgy
journal homepage: www.elsevier.com/locate/hydromet

Flow and mass transfer modelling for copper electrowinning:

development of instabilities along electrodes
Martin J. Leahy a,c,1, M. Philip Schwarz b,c,⁎
a
CSIRO Earth Sciences and Resource Engineering, Clayton, Victoria, 3168 Australia
b
CSIRO Computational Informatics, Clayton, Victoria, 3168 Australia
c
CSIRO Minerals Down-Under National Research Flagship, Australia

a r t i c l e i n f o a b s t r a c t

Article history: A computational fluid dynamics (CFD) model has been developed to simulate the copper electrowinning (EW)
Received 31 January 2014 process, and applied to model the flow and mass transfer in the inter-electrode gap for a single plate pair, with
Received in revised form 10 April 2014 geometrical and operation parameters typical of industrial EW operation. The CFD model predicts a recirculation
Accepted 20 April 2014
zone in all cases, driven by oxygen bubbles rising along the anode, with the electrolyte deflecting at the upper free
Available online 26 April 2014
surface and recirculating down to the base of the electrode space. The CFD model results showed that laminar
Keywords:
natural convection driven by concentration-related density deficiency is dominant along the lower part of the
CFD cathode. Strong eddies arise along the cathode where copper depletion becomes large enough to drive buoyancy
Copper Electrowinning instabilities: the instability is analogous to the waves formed in natural convection on a vertical heated plate for
Buoyant instability Prandtl number much greater than 2. Limiting current density can be increased by decreasing the boundary layer
Two phase flow thickness which can be achieved by increasing the velocity past the cathode, or by triggering flow instabilities
Mass transfer such as the buoyancy-generated ones described above or more conventional shear-driven instabilities. Higher
up the cathode, the natural convection profile becomes completely broken up by the recirculating down-flow.
Similar instabilities also form close to the anode due to build-up of oxygen bubbles, the fluctuating velocities
associated with the anode instabilities being much higher than those at the cathode.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction can elucidate the factors affecting the transport rate of copper to the
cathode surface, which is a major rate limiting factor in plating of copper
Copper electrowinning (EW) is the process of depositing solid cop- on the cathode.
per onto a cathode from a leach solution by passing an electric current There are only a limited number of published studies of the hydrody-
through the electrolyte. Industrial copper EW is generally carried out namics of the electrolyte in electrowinning cells. Ziegler (1984) and
in large rectangular cells, in which numerous alternating cathode and Ziegler and Evans (1986) collected limited data of the vertical velocity
anode plates are stacked vertically: current passing between the elec- profile in an industrial scale system (94 cm × 12.5 cm), and compared
trodes causes copper ions to deposit at the cathode, whilst oxygen bub- the velocity profiles with a simple fluid dynamics model with some suc-
bles are generated on the anode. It is well known that the rising oxygen cess. They found that a large scale recirculation is driven by oxygen bub-
bubbles cause a large recirculation zone to develop in the space be- bles rising at the anode. Turbulence intensity was predicted based on
tween the electrodes and that this recirculation has a strong effect on applying a two-equation turbulence model to this time-averaged
mass transfer to the cathode, due to the mixing generated by the recir- flow: the region very near the anode is thus predicted (and measured)
culation (Graydon and Kirk, 2001). A computational fluid dynamics to have very high levels of turbulence, with turbulence intensity of
(CFD) model of the bubble generated recirculation can allow insights order 50–100% being seen in both measurements and predictions.
into the hydrodynamic behaviour of electrolyte in the spaces between High levels of turbulence were also predicted to occur at the electrolyte
plates, and can in this way assist in understanding the details of copper surface where the upward stream is turned towards the cathode and at
mass transfer to different parts of the cathode. In particular, a CFD model the top of the cathode itself. The model was also applied to an experi-
mental arrangement for which mass transfer measurements had been
made at the cathode (Ettel et al., 1974): mass transfer coefficient pre-
⁎ Corresponding author at: CSIRO Process Science and Engineering, Australia. Tel.: +61
dicted by the CFD turbulence model agreed with the measurements at
3 95458898.
E-mail addresses: [email protected] (M.J. Leahy),
the bottom of the cathode (approximately 1.3 × 10− 6 ms− 1) and
[email protected] (M.P. Schwarz). were similar at the top of the cathode (2.2–2.6 × 10−6 ms−1), but the
1
Present address: Wood Group Kenny. detailed variation with height was not well reproduced.

http://dx.doi.org/10.1016/j.hydromet.2014.04.010
0304-386X/© 2014 Elsevier B.V. All rights reserved.
42 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53

The papers of Filzwieser et al. (1999) and Filzwieser (2000) describe the mesh maintains good discretisation accuracy. Compute times were
an extensive experimental program of detailed measurements of the typically less than a day running in serial mode. Three dimensional sim-
velocity profiles in an EW cell, as required for reliable CFD validation. ulations required several weeks.
This experimental data has been used by Leahy and Schwarz (2010) A two-phase Eulerian–Eulerian (“two-fluid”) gas liquid CFD model is
to validate a detailed CFD model. Filzwieser et al. (1999), Filzwieser employed in ANSYS CFX (2007), with the liquid phase (electrolyte)
(2000), and Filzwieser et al. (2002) discuss some aspects of the inter- treated as the continuous phase, and the oxygen bubbles as the dis-
electrode hydrodynamics, including the basic recirculation zone that persed phase. The technique models the dispersed phase as a quasi-
develops in the inter-electrode gap, and the variation in the local copper continuum, and has been successfully applied to bubble plumes and
concentration close to the cathode, as a function of height along the the recirculation driven by the plumes (Schwarz, 1996). We assume
cathode. The authors identify vertically rising flow in the lower part of mono-sized bubble distribution of diameter 150 micron. This estimate
the cell very close to the cathode, which they interpret as being natural is based on reports that the bubbles generated are fine with a distribu-
convection driven by the buoyancy of electrolyte depleted of copper, the tion from 50 micron to 200 micron (Filzwieser, 2000; Ziegler and
dominant flow mechanism in electrorefining (Leahy and Schwarz, Evans, 1986) — the predicted flow in the rising bubble plume is insensi-
2007, 2011). (This flow was also seen by Ziegler and Evans (1986), tive to the exact bubble size provided the terminal (slip) velocity is
but not mapped in detail.) A qualitative description is also given of the small relative to the plume velocity.
dynamic behaviour of the flow between the electrodes — the complex- A mass balance equation is solved, which ensures conservation of
ity of the fluctuating flow suggests that mass transfer will not be fully mass of each phase. This is known as the equation of continuity, and
understood with a time-averaged approach. In this paper we analyse for each phase is given by
the flow dynamics in more detail, especially as it relates to mixing and
transport, using geometrical and operational parameters typical of in- ∂ðα i ρi Þ
dustrial cells. þ ∇  ðα i ρi vi Þ ¼ Si ð1Þ
∂t
The paper has the following structure: a description of the CFD
model is given followed by the detailed description of the results, and
where for phase i (i =1 is liquid and i =2 is gas), Si is the mass source/
conclusions.
sink term (for example at the anode and free surface where gas enters
and leaves), ρi is the phase density, and vi is the velocity. A momentum
2. Electrowinning CFD model equation known as the Navier-Stokes equations is solved for each
phase, which balances the forces present in the two-phase flow. The
2.1. Model structure and equations Navier-Stokes equation in vector form is given by

The CFD electrowinning model is two dimensional (2D) in the Y-Z

∂ðα i ρi vi Þ 0
plane, and is applied to a cross section of the cell, as shown in Fig. 1 þ ∇  ðα i ρi vi ⊗vi Þ ¼ −α i ∇p þ ∇
∂t h

i
with the assumption that the flow is uniform in the third (horizontal) T
 α i μ i þ μ T;i ∇vi þ ð∇vi Þ þ Mi
dimension (X direction), parallel to the electrodes. The rectangular ge-
þ Si vi ð2Þ
ometry of the gap region allows the use of an orthogonal mesh, as illus-
trated in Fig. 1. The typical number of cells in the Y and Z directions is
100 and 200 respectively, with a geometric progression of cell widths where p′ is the (modified) pressure, and Mi is the sum of the body forces,
in the Y direction ensuring that the cells near the electrodes are of described below. The laminar viscosity is denoted μi (kg·m−1 s−1), and
order 20 μm wide. The height of cells is also refined in the region near μT,i (kg·m−1 s−1) is the turbulent viscosity, described below. The tran-
the liquid surface to approximately 1 mm. The high aspect ratio of the sient terms in Eqs. (1) and (2) are not included when the model is run
gap geometry causes high cell aspect ratio, but the orthogonality of as a steady state.

Fig. 1. Schematic diagram of the CFD geometry — side, cross section and cross section mesh views. A much larger number of cells than shown is actually used in the CFD model.
 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53 43

The sum of the body forces is given by Navier-Stokes (RANS) methods. However RANS models, such as the
SST turbulence model, have the advantage of being able to be imple-
Mi ¼ Bi þ Fi þ Ti þ Ai ð3Þ mented on a reasonably coarse mesh, compared to the large-eddy sim-
ulation (LES) turbulence approach. Although the LES technique resolves
where Bi, Fi, Ti and Ai are the phase-related buoyancy force, bubble drag
the turbulent eddy structures more accurately, it requires significantly
force, turbulent dispersion force and concentration-related buoyancy
more computational resources than two equation turbulence models,
force, respectively. The forces are given by
as it needs to be solved transiently in 3D and with a fine mesh. For
B2 ¼ −B1 ¼ α i ðρi −ρref Þg Phase−related buoyancy force ð4Þ this reason, a RANS approach is used in the present work.
The drag coefficient is dependent on the bubble size and, for small
bubbles, this is well described by the Schiller-Naumann equation:
3 CD
F2 ¼ − F1 ¼ − ρ α jv −v1 jðv2 −v1 Þ Drag force ð5Þ 24

4 d 1 2 2 CD ¼ 1 þ 0:15Reb
0:687
ð10Þ
Reb

T2 ¼ −T1 ¼ C td ρ1 k∇α 2 Turbulent dispersion force ð6Þ where Reb (−) is the bubble Reynolds number given by

dρ1 jv2 −v1 j

Reb ¼ ð11Þ
A1 ¼ α 1 ½−ρ1 gβðC−C ref Þ; A2 μ1
¼ 0 Concentration−related buoyancy force liquid ði ¼ 1Þ ð7Þ
where d (m) is the bubble diameter. For 150 μm diameter bubbles, as
where g (m·s− 2) is the gravity vector, v2 − v1 (m·s− 1) is the slip noted in the experiment by Filzwieser et al. (1999) and Filzwieser
velocity, |v 2 − v 1 | is the size (modulus) of the slip velocity, ρ ref (2000), the bubble Reynolds number is Reb ~ 0.3 causing CD (from
(kg·m− 3 ) is the reference density taken as that of the electrolyte Eq. (10)) to be very large, which means the bubble motion is in the
(i = 1), CD is the drag coefficient, Ctd is the turbulent dispersion co- Stokes regime where the drag is sufficiently high for the two phases to
efficient (taken as 1), k (m2·s− 2) is the turbulence kinetic energy, C have almost the same velocity.
(kg·m− 3) is the concentration of copper, Cref (kg·m− 3) is the ref- A closure equation is required for the volume fraction equations, and
erence concentration of copper (initial and inlet concentration), is given by
and β (m 3 ·kg − 1 ) is the coefficient of expansion for the copper
species. α1 þ α2 ¼ 1 ð12Þ
The turbulent viscosity in Eqs. (2) and (13) is determined by solving
turbulence transport equations using the well known k-ω model The additional transport equation for the copper species (Cu2+) in
(Wilcox, 1986). This model was originally derived for single phase the liquid phase is given in steady state by
flows, but can be used for multi-phase flows by solving the k-ω model
    
for the continuous liquid phase, and using the same turbulence quanti- μ T;1 C
∇  ðα 1 Cv1 Þ ¼ ∇  α 1 ρ1 D þ ∇ þ SCu ð13Þ
ties (k, ω) for the dispersed gas phase. The gas turbulent viscosity (μT,2) ScT ρ1
for the gas phase is taken to be:
where SCu (kg·m−2 s−1) is the source term, which describes the flux of
ρ μ T;1 copper at the cathode, or the source or sink of copper at the inlet and
μ T;2 ¼ 2 ð8Þ
ρ1 σ T outlet, respectively, D (m2 s− 1) is the diffusion coefficient of copper
ions, and ScT (−) is the turbulent Schmidt number, which is typically
where the turbulent Prandtl number σΤ is taken as 1 since the bubble re-
given the value 0.9, as assumed in this work. This equation can be de-
laxation time is short compared to turbulence dissipation time scales.
rived in exact form for a binary electrolyte by eliminating the term for
In this work several turbulence models were tested (e.g., k-ε, stan-
the electric field force and applying the condition of charge neutrality
dard k-ω and Shear Stress Transport (SST)) but the SST turbulence
(Newman and Thomas-Alyea, 2004). For the multi-component electro-
model was found to give the best velocity profile prediction close to
lyte in an electrowinning cell, the equation will be approximately valid
the wall. The SST turbulence model uses a combination of the k-ω and
if the concentration of other components is low.
k-ε models, via a blending function which switches between the two
We define the turbulence intensity, IT, as the ratio (expressed as a
models near to the walls: for this reason the SST model is known to be
percentage) of the root-mean-square (RMS) turbulence velocity fluctu-
a useful two equation turbulence model. The turbulent viscosity
ation (v′) to the time-averaged speed Vav at a point near the electrodes
(Menter, 1993) can be written in terms of the transported variables —
(typically at the point where the vertical velocity is a maximum):
kinetic energy, k, and eddy frequency, ω, as follows:
0
αk v
ν T;1 ¼ ð9Þ IT ¼ 100  ð14Þ
maxðαω; S F 2 Þ V av

where νT,1 is the turbulent kinematic viscosity (m2 s−1) given by νT,1 =
μT,1/ρ1, and where F2 is a blending function which limits the turbulent 2.2. Model boundary conditions
viscosity in the wall boundary layer, where conventional turbulence
models are invalid. S is an invariant measure of the strain rate, and α = 2.2.1. Boundary conditions on anode
5/9 is a dimensionless parameter. The geometry shown Fig. 1 is useful to describe the boundary condi-
As will be seen, the flow in the inter-electrode gas is predicted to be tions used. On the anode side where oxygen is produced, the boundary
unsteady, with the unsteadiness arising from interactions between condition for oxygen gas source is based on Faraday's Law of electroly-
buoyancy (density) fluctuations and flow. Since these velocity fluctua- sis. If we assume 100% current efficiency, the superficial gas production
tions will in reality interact with turbulence fluctuations arising from rate (volume of gas generated per unit anode surface area) is given by:
shear instabilities, the flow would be most accurately modelled using
an eddy-resolved method such as direct numerical simulation (DNS) 1 IRT
v2;Y ¼ ð15Þ
or large eddy simulation (LES) rather than with Reynolds averaged 4 P atm F
44 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53

where R (J·K−1 mol−1) is the gas constant, T is the temperature (K), Table 1
Patm (Pa) is atmospheric pressure, F (A·s·mol−1) is Faraday's constant Table of parameters in EW model.

and I (A·m−2) is the current density, which is assumed uniform over Parameter Value
the surface of the anode: in future work this assumption could be
Current Density I (A m−2) 300
addressed to include a non-uniform current density. Temperature (°C) 50
The oxygen mass flow rateṁoxygen can be calculated based on the Superficial Flow Velocity (m s−1) 2.5 × 10−2
superficial velocity from Eq. (15) Liquid Laminar Viscosity μ1 (kg/m/s) 1.18 × 10−3
Liquid Density ρ1 (kg m−3) 1192.4
Oxygen Density ρ2 (kg m3−) 1.21
ṁoxygen ¼ v2;Y Aan ρ2 ð16Þ
Oxygen Viscosity μ2 (kg/m/s) 2.18 × 10−5
Diffusion Coefficient D (m2 s−) 8.62 × 10−10
where Aan (m2) is the area of the anode. Coefficient of Expansion β (m3 kg−1) 0.0019 (copper)
Reference Concentration Cref (kg m−3) 40 (0.6294 M copper)
2.2.2. Boundary conditions on cathode for copper Acid Concentration (kg m−3) 180 (1.853 M acid)
Molecular Weight MCu (kg mol−1) 0.063546 (copper)
The flux of copper at the cathode m• Cu (kg m− 2 s− 1) based on
Height of Electrode (mm), H 1000
Faraday's Law is as follows: Width of Electrode Gap (mm), h 40

• I MCu NB viscosity, density, diffusion coefficient of expansion values are based on a source in the
mCu ¼ − ð17Þ literature, or tables of data from Zaytsev and Aseyev (1992).
zF 1000

where I (A m−2) is the current density, F (A s mol−1) is Faraday's con-

stant, z (−) is the valency (in this case equal to 2), and MCu (g mol−1)
is the molecular weight of copper. The boundary condition for Eq. (12) of variables in the bulk is small. Incoming electrolyte enters from the
at the cathode walls is essentially the diffusional flux of copper ions right on the anode side. The figures, (a)–(f), show the velocity vector
which, for a binary electrolyte or a dilute solution of copper in a sul- field, contours of copper concentration, oxygen volume fraction, turbu-
furic acid supporting electrolyte, can be shown to be approximately lence eddy viscosity ratio, speed and streamlines, respectively. An over-
(Newman and Thomas-Alyea, 2004) all recirculation zone develops, due to the bubbles rising from the anode
and dragging electrolyte upwards. At the top, the electrolyte turns back

• I 1−t þ M Cu downwards and returns to the anode. There is an intense recirculation
mCu ¼ ð18Þ at the top of cell, with a depth roughly the same as the gap width.
zF 1000
Below that is a zone of lower speed downward flow.
Here t+ (−) is the transference number, defined as the proportion of The copper concentration contours plotted in Fig. 2(b) show that the
current carried by copper ions in a uniform solution without concentra- concentration over the bulk of the inter-electrode gap is uniform. Re-
tion gradients (Newman and Thomas-Alyea, 2004). gions of lower concentration can be seen very close to the cathode
and these will be examined in more detail later in the paper.
2.2.3. Wall and free surface boundary conditions From the oxygen volume fraction contours shown in Fig. 2(c), it is
At all walls, no-slip boundary conditions are applied for the liquid evident that bubbles are re-entrained in the recirculation, due to the
phase, and for the gas phase, free-slip boundary conditions are applied. small size of the bubbles. The oxygen volume fraction is around 2.5%
At the free surface, a free-slip (no friction) boundary condition is applied at the top of the inter-electrode gap, and gradually decreases with in-
to the liquid phase, whilst for the gas phase, a degassing boundary con- creasing depth in the cell to negligible values at around mid-height.
dition is used. The degassing boundary condition allows gas bubbles to This distribution is set up because the bubbles are dragged downwards
leave the liquid through the surface at the rate at which they arrive at to a point where the down-flow velocity is just high enough to over-
the surface. At the walls, the grid resolution is such that Y+ = 1, and come the (upwards) buoyancy force; i.e., the downward electrolyte ve-
therefore in the k-ω formulation integration of Eq. (3) in the CFD solver locity equals the upwards bubble slip (terminal) velocity. The bubble
is carried out to the wall and wall functions are not required. size used here is 150 μm; a different assumed bubble size would result
in gas being entrained to a somewhat different depth.
2.2.4. Inlet and outlet boundary conditions Fig. 2(d) shows that the turbulence eddy viscosity ratio (turbulent to
An inlet and outlet arrangement is added near the base of the geom- laminar viscosity), which is a measure of turbulence intensity, is highest
etry to allow through-flow and the introduction of copper to avoid con- near the anode plate (where rising bubbles result in significant velocity
tinuous depletion. The presence of an inlet and outlet is similar to the shear in the electrolyte) and towards the top of the plate pair where the
industrial scenario where there is a gap at the base of the electrodes. strong recirculation zone occurs. These higher velocity zones are seen in
the speed plot shown in Fig. 2(e). The eddy viscosity ratio is lowest in
3. Results of CFD electrowinning model the lower parts of the gap, and near the cathode.
Fig. 3 shows close-up views of the near-cathode regions in the lower,
3.1. Model parameters middle and upper zones of the cell, where we define Z′ as the height
above the base of the electrodes (Z′ = Z − 100 mm). The velocity vector
The computational mesh used for the simulation has a cell spacing plot in Fig. 3(a) shows that for Z′ b 400 mm, natural convection occurs,
which is finer near the walls than in the middle, so as to resolve the which is because the regions close to the cathode become depleted in
higher velocity gradients near the walls. The parameters used in the copper and hence are lighter than the surrounding bulk electrolyte.
CFD model, including boundary conditions, are given in Table 1. Geo- Buoyancy forces cause electrolyte in these regions to rise. In electro-
metrical and operational parameters are taken to be typical of industrial refining, this phenomenon (together with the inverse sinking at the
cells. anode caused by higher density copper enriched electrolyte) dominates
the flow in the inter-electrode gap; in EW this phenomenon generally
3.2. Discussion of CFD results only occurs very close to the cathode (closer than 1 mm) where the gra-
dients are highest, and only in the lower part of the cell.
The results shown in Fig. 2(a)–(f) are for a 2D cross section with the Above Z′ = 300–400 mm, eddies begin to develop along the cathode
anode on the right (high Y) and the cathode on the left (low Y). The sim- at various positions (Fig. 3(b)) as the downward flow from higher on
ulation is run to a quasi-steady state, at which the variation in the values the cathode interacts with the upward natural convection from lower
 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53 45

Fig. 2. (a) Electrolyte velocity vector field, (b) contours of copper concentration (kg m−3), (c) contours of oxygen volume fraction (−) (log scale), (d) contours of eddy viscosity ratio (−)
(log scale), (e) contours of speed (mm s−1) (log scale), and (f) streamlines of electrolyte. All plots stretched by factor 3 in Y direction.

on the cathode. The copper distribution at mid-height in Fig. 3(b) shows from the copper gradient near the cathode (in the layer within around
regions of low concentration moving away from the cathode by the ad- 1 mm): the electrolyte is dragged upwards in a stream of buoyant elec-
vective action of the eddies. trolyte and in this stream the electrolyte becomes increasingly depleted
In Fig. 3(c) we see a close-up view of the top region, showing the in copper as it rises. Note that, although the concentration gradient at
natural convection eddies surrounded by the down-coming electrolyte, the cathode wall remains the same, as required by constant current den-
which brings higher copper concentration solution from the bulk to- sity, the overall difference between the concentrations near the wall and
wards the cathode wall. Meanwhile, the eddy shown in Fig. 3(c) dis- in the bulk (defined as ΔC) increases as the electrolyte moves upward.
perses the low copper concentration solution which develops in the Eventually ΔC becomes so large that the buoyancy force causes a flow
low turbulence natural convection up-flow, and mixes it into the bulk. instability to form. The instabilities begin to occur at around Z′ =
300 mm, and become more severe as the electrolyte moves into the
3.3. Transient eddy structure close to cathode middle region.
In the middle region, the instabilities move away from the cathode
The model shown in Fig. 3 has been run to a steady state, and these wall and then interact with the hydrodynamics in the bulk, which is
results were used to restart a transient simulation with short timestep, predominantly down-moving bubble-driven recirculation flow. The
to study the temporal evolution of the eddies discussed. Instantaneous bulk down-flow may tend to amplify the instability somewhat. Howev-
snap shots of the eddies evolving along the cathode are shown in er, moving further up towards the top third (high region), the instabil-
Figs. 4 and 5: the points M1, M2, etc correspond to each eddy in the ities tend to be dissipated by the down-coming recirculation: this is
mid region (M), and the H1 and H2 correspond to each eddy in the because the recirculating down-flow flow is stronger in the upper re-
high region (H). This numbering was then to keep track of the eddy gions of the inter-electrode gap than lower down.
movements over time. In the lower regions in Figs. 4 and 5 we see the Elder (1965a,b) published high quality photographic snapshots of
region Z′ b 300 mm does not change in time, and is apparently laminar the flow along the heated plates that show small unstable ‘hooks’,
and steady. However, for Z′ N 350 mm, eddies form and move upwards which grow until they ‘break’ causing turbulent motion. The growth
along the cathode, eventually moving out into the bulk inter-electrode comes from increased local heat transfer due to the vortex itself, causing
region and being dissipated by the large-scale down-flow. Immediately unstable growth of density gradient. These hooks appear to be due to
after the eddies form, the copper concentration increases as the surface the same instability phenomena as that causing the development of
is refreshed to some extent with electrolyte from the bulk. The reason the eddies shown in Fig. 3: the analogy between heat transfer and
the eddies form can be understood by examining the flow behaviour mass transfer implies that similar instabilities may occur in the near-
close to the cathode: we examine the flow close to the cathode at cathode region as occurs near a vertical heated plate.
three different height ranges — near the bottom, at mid-cell and in the Jannsen and Armfield (1996) have carried out stability analysis for
upper region. the instability on a vertical heated plate, and show that the nature of
In the bottom third (approximately Z′ b 300 mm) the flow is laminar the perturbations changes at Prandtl number (Pr) of 2: for Pr less than
and the electrolyte is dragged upwards by the buoyancy forces that arise 2, the unstable waves are shear driven (albeit with the shear generated
46 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53

Fig. 3. Close view of velocity vector field (left) and copper contours (kg m−3) (right) at three heights: (a) low Z′ position (Z′ b 300), showing laminar region, (b) mid Z′ position
(350 b Z′ b 600) showing turbulent eddies forming and refreshing the cathode surface and (c) high Z′ position (Z′ N 650 mm) showing region where eddies are dissipated by the
down-coming bubbly flow. All plots stretched by factor 3 in Y direction.

by buoyancy), whereas for Pr greater than 2, the unstable waves are the cathode instabilities should be directly buoyancy driven. This is ex-
directly buoyancy driven. When the thermal diffusivity is much larger actly what is seen in the CFD model (Figs. 3, 4 and 5). The experiments
than the momentum diffusivity, perturbations cannot build up in the conducted by Elder (1965a,b) were also in the regime where Pr is much
temperature distribution near the plate — enthalpy diffuses away greater than 2, as expected since the instabilities are clearly buoyancy
from the plate, driving a high velocity shear, which is then unstable to generated.
waves because of the relatively lower momentum diffusivity (i.e., vis- An important result of the instabilities discussed here is that they in-
cosity). The analogous dimensionless number to Prandtl number for crease the transport rate of heat and mass to (or from) the surface.
mass transfer is Schmidt number (Sc), defined as the ratio of momen- Evans and Greif (1997) show that the buoyancy-driven instability can
tum diffusivity and mass diffusivity. One would expect a similar change increase Nusselt number in one case to 3.7, 4.5 and 4.9 for three increas-
in the nature of the instability at a Sc of 2. Using the diffusivity of copper ing values of Grashof number (i.e. increasing buoyancy relative to con-
of 8.62 × 10−10 m2 s−1 and kinematic viscosity (momentum diffusivity) vection), where Nusselt number in this case was defined in terms of
of 9.9 × 10−7 m2 s−1 gives a Schmidt number of 1150, suggesting that the normal temperature gradient at the surface, so that it equals unity
 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53 47

Fig. 4. Time snap shots of close-up view of velocity vector field and copper contours (kg m−3) showing eddies evolving along the cathode at three heights: low, mid and high. Shown at
three times (a) 8.4 seconds (b) 9.4 seconds and (c) 10.4 seconds (time snap shots are continued in Fig. 5). The horizontal scale is increased by a factor of 10 in Y direction.

for no forced convection. An analogous enhancement of Sherwood current density at which the surface concentration is zero is known as
number is to be expected, and is observed at the cathode in the present the limiting current, Ilim, since at this point, transport of ions from the
simulations of electrowinning. bulk cannot support a higher value:
As reviewed by Filzwieser et al. (2002), at regions very close to the
cathode surface where the velocity is vanishingly small, Fick's law can
zFD c0
be applied to reduce Eq. (17) to Ilim ¼  ð20Þ
1−t þ δ
 
zFD dc zFD c0 −cs
I¼ ¼ ð19Þ
1−t þ dx x¼0 1−t þ δ In industrial electrolysis, the technical current density is found to be
about 350 A cm−2, although the limiting current density can be as high
where δ is the boundary layer thickness, and c is the molar concentra- as 800 A cm−2 (Filzwieser et al., 2002).
tion, c = 1000C/MCu. (Note that the transport by current was omitted Since Eq. (17) is applied as a boundary condition in the CFD model,
in Filzwieser's paper.) As the applied current is increased, the surface the concentrations predicted in the simulations also satisfy Eq. (19).
concentration of copper cations, cs , can approach zero — the value of The copper transport rate predicted by the model can be used to
48 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53

Fig. 5. Time snap shots of close-up view of eddies evolving along the cathode at three heights low mid high, and at three times (a) 11.4 seconds (b) 12.4 seconds and (c) 13.4 seconds (time
snap shots continued from Fig. 4). The horizontal scale is increased by a factor of 10 in Y direction.

determine the limiting current density for any particular arrangement it is increased. Attempts to increase the current above this value re-
and for any location on the cathode using the relationship: sult in an uneven or nodular deposit which can trap impurities. The
limiting current can be increased by decreasing the boundary layer
δ ¼ D=kpot ð21Þ thickness, δ, which can be achieved by increasing the velocity past
the cathode, or by introducing flow instabilities such as the buoyancy-
where kpot is the maximum potential mass transfer coefficient for the generated ones described above or more conventional turbulence
particular flow conditions determined by the CFD model — this can be (shear driven instabilities) which disrupt, and thus effectively reduce
determined by applying a zero concentration boundary condition at the thickness of, the boundary layer. In practice, it is likely that introduc-
the cathode. tion of artificial velocity or concentration fluctuations at the base of the
The lower region of the cathode in which the flow is laminar, as cathode where electrolyte is being entrained into the boundary layer
shown in Fig. 3(a) is the first region to reach the limiting current as will trigger instabilities sooner, thereby increasing mass transfer and
 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53 49

Fig. 6. View close to anode (up to 7 mm away from anode) with vertical locations indicated as Z′ (height above base of electrodes). Note the horizontal scale is increased by a factor of 5.
Time snap shots of oxygen volume fraction (log scale), with velocity vectors and dots following the streamlines, showing eddies evolving along the anode around the start of the onset of
hooks. The times start from a reference time t0.
50 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53

Fig. 7. Full width cross section, and limited vertical view, with vertical locations indicated as Z′ (height above base of electrodes). Time snap shots of oxygen volume fraction (log scale),
with velocity vectors, showing eddies evolving along the anode. The times start from a reference time t0. The horizontal scale is increased by a factor of 5.

the limiting current in the critical zone at the lower part of the velocity is highest in the gas pocket, and this is shown by the correspon-
cathode. The practical feasibility of such an arrangement needs to be dence between the contour plot of the local Reynolds number in
investigated. Fig. 8(a), and the pocket of oxygen volume fraction (Fig. 8(b)) that is
being pushed away from the anode.
3.4. Transient eddy structure near the anode It appears as if the hooks along the anode arise due to the same phe-
nomena as those which form along the cathode, which were driven by
In Fig. 2 we showed the speed and oxygen volume fraction contour buoyancy forces arising from copper depletion, although the density
plots for the whole geometry. Figs. 6 and 7 show the velocity vector fluctuations are much greater in the bubbly region at the anode than
field and oxygen volume fraction for a small region near the anode — near the cathode. These hooks and the build-up of density gradients
Fig. 6 shows the region for Z′ between 280 mm and 350 mm, and and velocity gradient were discussed for electrorefining systems in
Fig. 7 shows the region for Z′ between 780 mm and 900 mm. Fig. 6 Leahy and Schwarz (2011). Fig. 9 shows results of the time-varying fluc-
reveals some small eddies (‘hooks’) – appearing to be small tuation of velocity at points on a vertical line 0.15 mm from the anode.
instabilities – forming along the anode around two thirds of the At these points near the anode there are fluctuations in velocity
anode height. The oxygen bubbles are pushed away from the anode as (reflecting the hooks which form quasi-periodically), which are large
buoyancy driven instabilities develop, as shown by the velocity relative to the time-averaged velocity. The quasi-periodic fluctuations
vectors becoming slightly off vertical away from the anode wall. This are much more frequent and larger in amplitude along the anode
phenomenon delivers a significant amount of gas into the bulk, increas- (Fig. 9) than the cathode (plotted at points with similar vertically spac-
ing the total gas hold-up in the bulk — as plotted in Fig. 2(c). Because the ing and 0.15 mm from cathode in Fig. 10), due to the larger instability
pocket of gas is pushed way from the anode wall, the gas pocket slows driving force (greater density gradient and velocity) at the anode. Ex-
down because it is removed from the fast upward moving plume: this is cept at a height of 99.5 cm, the time-varying velocity fluctuations
seen in the video animation (available online) of Fig. 6 by following the along the cathode in Fig. 10 are much more subtle than along the
dots (which represent particles following the streamlines) and observ- anode. Above 40 cm the fluctuation amplitude is around 1–2 mm s −1,
ing that the pocket of gas is pushed off and slows once it is out of the corresponding to relative fluctuation intensity of 1–2%. It is interesting
main upwards moving gas plume. The instability arises because in the that this is the same order of magnitude as the turbulence intensity es-
region along the anode, bubbles are produced and rise upwards, so timated by the RANS turbulence model applied in the CFD simulation: it
that the pocket of fluid builds up more gas until the volume fraction is is likely that the energy of the quasi-periodic fluctuations flows through
high enough (and consequentially the velocity is high enough) to to random turbulence at smaller length scales, and that these are
provide a sufficient buoyancy force for an instability to arise. The accounted for by the RANS model. Although turbulence is present
 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53 51

Fig. 8. Close-up view at one time step t0 1 mm from anode with vertical locations indicated as Z′ (height above base of electrodes), showing co-location of hook (instability) of (a) the local
Reynolds number and (b) oxygen volume fraction (log scale), with velocity vectors. The horizontal scale is increased by a factor of 5.

along both the cathode and anode, it develops more quickly along the is a two-phase (Eulerian-Eulerian gas-liquid) fluid dynamics model for
anode, and is maintained at a higher level than at the cathode, due to copper electrowinning in a single plate pair, which has been validated
the greater density gradient and velocity present along the anode. with experimental data by Leahy and Schwarz (2010).
(See Fig. 11.) Previously, very little has been known about the underlying fluid dy-
namics of electrolyte flow, bubble transport and copper distribution in
4. Conclusions an electrowinning cell, particularly in the space between electrodes.
This is despite the fact that fluid dynamics and mass transfer are well
A CFD model has been developed to simulate bubble-driven hydro- known to have a strong effect on the copper plating quality. This
dynamics and copper distribution in an electrowinning cell consisting of paper has established a detailed understanding of the bubble driven re-
a single anode–cathode pair. The model incorporates transport of oxy- circulation flow in the space between electrodes, including the copper
gen bubbles generated on the anode and copper in the electrolyte, and gradient driven buoyancy forces. The CFD model predicts a recirculation
includes the depletion of copper (and potentially other metal species in zone in all cases, due to the oxygen bubbles rising along the anode
other industrial processes) at the cathode and density-related buoyancy and dragging electrolyte upwards, which then turns at the top and
forces arising from metal concentration gradients. The model developed

Fig. 9. Vertical velocity (mm s −1) versus time at vertically points close to the anode Fig. 10. Vertical velocity (mm s −1) versus time at vertically points close to the cathode
(0.15 mm offset). (0.15 mm offset).
52 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53

Notation
Ai Concentration related buoyancy force of phase i, N m−3
Aan Area of anode, m2
Bi Phase-related buoyancy force for phase i, N m−3
C Copper concentration, kg m−3
Cref Reference concentration of copper, kg m−3
ΔC Difference in bulk to wall concentration of copper, kg m−3
CD Drag coefficient, −
Ctd Turbulent dispersion coefficient, −
d Bubble diameter, m
D Diffusion coefficient, m2 s−1
F Faraday's constant, A s mol−1
Fi Drag force of phase i, N m−3
F2 Blending function, −
g Gravitational acceleration vector, m s−2
H Height of electrode, mm
h Width of electrode gap, mm
Fig. 11. Turbulence intensity (%) along lines close to anode (solid) and cathode (dotted) at
I Current density, A m−2
one instant in time, indicating developing turbulence along cathode and anode. IT Turbulence intensity, −
k Kinetic energy of turbulence, m2 s−2
kpot Maximum mass transfer coefficient, m s−1
recirculates down to the base of the plate pair. Laminar natural convec- m• Cu Flux of copper at the cathode walls, kg m−2 s−1
tion is dominant at the base of the cathode, but becomes disturbed with m• oxygen Flux of oxygen at the anode and cathode walls, kg m−2 s−1
increasing height by unstable eddies developing along the cathode. MCu Molecular weight of copper, g mol−1
These cathode instabilities form and move into the bulk, thus making Mi Sum of body forces of phase i, N m−3
the flow in the bulk more unstable, and improving mixing of copper Patm Atmospheric pressure, Pa
to the cathode. Using values for the diffusivity of copper and electrolyte p Pressure, Pa
kinematic viscosity gives a Schmidt number of 1150, suggesting that the p′ Modified pressure, Pa
cathode instabilities should be directly buoyancy driven, similar to the Pr Prandtl number, −
instabilities seen in heat transfer to a vertical heated plate when the R Gas constant, J/K/mol
Prandtl number is greater than 2. These instabilities are different from Reb Bubble Reynolds number, −
the more conventional shear-driven instabilities that are seen for ScT Turbulent Schmidt number, −
Prandtl number less than 2, and are related to the fluctuations that S Invariant measure of the strain rate, s−1
occur in the transition to conventional turbulence. The limiting current Si Mass source/sink, kg m−2 s−1
can be increased by decreasing the boundary layer thickness, δ, which can t+ Transference number, −
be achieved by increasing the velocity past the cathode, or by introducing t0 Initial reference time for transient simulations, s
flow instabilities such as the buoyancy-generated ones described Ti Turbulent dispersion force of phase i, N m−3
above or more conventional turbulence (shear driven instabilities) T Temperature, K
which disrupt, and thus effectively reduce the thickness of, the vi Velocity vector of phase i, m s−1
boundary layer. In practice, it is possible that introduction of artificial v′ RMS turbulence velocity fluctuation, m s−1
velocity or concentration fluctuations at the base of the cathode Vav Time-averaged vertical velocity, m s−1
where electrolyte is being entrained into the boundary layer may X X direction coordinate, m
trigger instabilities sooner, thereby increasing mass transfer and Y Y direction coordinate, m
the limiting current in the critical zone at the lower part of the cath- z Valency, −
ode. Higher up the cathode, the natural convection profile becomes Z Z direction coordinate, m
completely broken up by the strength of the downflow from the bubble Z′ Effective height above base of electrodes, m
driven recirculation.
Along the anode, gas volume fraction builds up and develops into
fast moving pockets of gas, where an instability develops either at a Greek symbols
critical velocity and/or a critical density difference with the bulk αi Volume fraction of phase i
electrolyte density. The pocket is then pushed off the anode into α Dimensionless turbulence parameter, −
the bulk, and this phenomenon delivers a significant amount of gas β Coefficient of expansion, m3 kg−1
into the bulk, increasing the overall gas hold-up in the bulk. Turbu- δ Boundary layer thickness, m
lence is present along both cathode and anode, but develops more ε Dissipation rate of turbulence energy, m2 s−3
quickly along the anode and is maintained at a higher level than at μi Liquid laminar dynamic viscosity, kg m−1 s−1
the cathode due to the greater density and velocity present along μΤ ,i Turbulent dynamic viscosity, kg m−1 s−1
the anode. νT Turbulent kinematic viscosity, m2 s−1
In future work, experimental work should be undertaken to investi- ρι Density of phase i, kg m−3
gate and validate the CFD-predicted eddies which form along the cath- σT Turbulence Schmidt number, −
ode and anode: in particular the eddies which form along the anode, ω Eddy frequency, s−1
and their associated ejections of oxygen bubbles into the bulk, and the
interplay between eddies forming along the cathode and the down-
coming recirculating zone fluid. Furthermore, future work should aim Subscripts
to develop a comprehensive understanding of the role of the transition atm atmospheric
from laminar to turbulent flow along the cathode, and its relation to b bubble
copper electrowinning plating quality. Cu Copper
 M.J. Leahy, M.P. Schwarz / Hydrometallurgy 147–148 (2014) 41–53 53

i Phase i (i = 1 water, i = 2 gas) Filzwieser, A., 2000. Modellierung der kathodennahen Vorgange in der Kupferelektrolyse
(In German) — translated “Modelling the near-cathode processes in copper electrolysis”.
ref Reference (PhD thesis) Leoben University, Austria p. 167.
T Turbulent Filzwieser, A., Hein, K., Hanko, G., Grogger, H., 1999. Application of two phase hydrody-
namics modelling to an electrowinning cell. Proc. Copper-Cobre. Electrorefining and
Electrowinning of Copper, 99 vol. III, pp. 695–709.
Filzwieser, A., Hein, K., Mori, G., 2002. Current density limitation and diffusion boundary
Superscripts layer calculation using CFD method. J. Met. 28–31 (April).
T Transpose Graydon, J., Kirk, D., 2001. Suspension codeposition in electrowinning cells: the role of
hydrodynamics. Can. J. Chem. Eng. 69, 564–570.
′ Modified pressure p′ or average velocity u′ Jannsen, R., Armfield, S., 1996. Stability properties of the vertical boundary layers in
′ Used for effective height Z′ above base of electrodes differentially heated cavities. Int. J. Heat Mass Transfer 17, 547–556.
Leahy, M., Schwarz, M.P., 2007. Computational fluid dynamics modelling of natural
convection in copper electrorefining. 16th Australasian Fluid Mechanics Conference,
Gold Coast, Australia, December 2007.
Leahy, M., Schwarz, M.P., 2010. Experimental validation of a computational fluid
Acknowledgements dynamics model of copper electrowinning. Metall. Mater. Trans. B 41, 1247–1260.
Leahy, M., Schwarz, M.P., 2011. Modeling natural convection in copper electrorefining:
The reported research was stimulated by work originally funded by describing turbulence behavior for industrial-sized systems. Metall. Mater. Trans. B
42B, 875–890.
AMIRA P705A sponsors. The authors also gratefully acknowledge Peter Menter, F.R., 1993. Zonal two equation k-ω turbulence models for aerodynamic flows.
Witt, Darrin Stephens and Graeme Lane for helpful CFD assistance and AIAA Paper 93-2906.
discussions. Newman, J., Thomas-Alyea, K., 2004. Electrical Systems, 3rd ed. John-Wiley and Sons,
New Jersey p. 14.
Schwarz, M.P., 1996. Simulation of gas injection into liquid melts. Appl. Math. Model. 20,
41–51.
References
Wilcox, D., 1986. Multiscale model for turbulent flows. Proceedings of the 24th AIAA
Aerospace Sciences Meeting. American Institute of Aeronautics and Astronautics,
ANSYS, 2007. CFX-11 Solver. ANSYS Inc., Canonsburg, USA (website address
Reno, NV, USA, pp. 1311–1320 (January).
www.ansys.com/cfx).
Zaytsev, I., Aseyev, G., 1992. Properties of aqueous solutions of electrolytes. CRC Press Inc.,
Elder, J.W., 1965a. Laminar free convection in a vertical slot, Part I. J. Fluid Mech. 23,
America.
77–98.
Ziegler, D., 1984. A study of electrowinning and electrorefining cell hydrodynamics, PhD.
Elder, J.W., 1965b. Turbulent free convection in a vertical slot, Part II. J. Fluid Mech. 23,
University of California, Berkeley.
99–111.
Ziegler, D., Evans, J., 1986. Mathematical modelling of electrolyte circulation in cells with
Ettel, V.A., Tilak, B.V., Gendron, A.S., 1974. Measurement of cathode mass transfer
planar vertical electrodes: II electrowinning cells. J. Electrochem. Soc. 133 (3),
coefficients in electrowinning cells. J. Electrochem. Soc. 121, 867–872.
567–576.
Evans, G., Greif, R., 1997. Buoyant instabilities in downward flow in a symmetrically
heated vertical channel. Int. J. Heat Mass Transfer 40, 2419–2430.