Minerals Engineering 179 (2022) 107447

Contents lists available at ScienceDirect

Minerals Engineering
journal homepage: www.elsevier.com/locate/mineng

Full-Scale trial of the REFLUX™ flotation cell

S.M. Iveson *, J.L. Sutherland, M.J. Cole, D.J. Borrow, J. Zhou, K.P. Galvin
ARC Centre of Excellence for Enabling Eco-Efficient Beneficiation of Minerals, Newcastle Institute for Energy and Resources, University of Newcastle, Callaghan NSW
2308, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: A 2 m diameter REFLUX™ Flotation Cell was fed at roughly 210 m3/h, equivalent to a flux of 1.9 cm/s, about
REFLUX™ Flotation Cell twice the maximum rate used in conventional flotation cells. The coal feed slurry had 59–64 wt% head ash and
Full-scale nominal size range − 0.100 mm (Sauter mean size 0.004 mm). The air, wash water and underflow rates were 180
Scale up
m3/h, 65 m3/h and 235 m3/h respectively, giving a positive downwards wash water bias flux of 0.2 cm/s.
Oil agglomeration
Product ashes of 11–15 wt% were obtained at combustible recoveries of 58–75 %, with results on or better than
the tree curve. These initial results demonstrate that the beneficial hydrodynamics seen at laboratory scale are
realised at full-scale. Also demonstrated is the use of a novel oil-agglomeration technique to obtain detailed
performance versus size data, showing that high hydrophobic recoveries were being obtained at sizes down to
0.001 mm.

1. Introduction entrainment of bubbles in the underflow tailings stream, and thus a loss
in recovery of the hydrophobic particles. All conventional flotation
Since its introduction in 1905, flotation has become one of the most machines are constrained by the requirement for a distinct upper froth
widely used beneficiation technologies in the minerals processing in­ layer to form above the bubbly mixture (pulp) zone below. At the
dustry, while also finding application in a wide range of other industries. “flooding condition”, defined here when the demarcation between these
Flotation is often the only viable method for separating complex ores zones disappears, control of conventional cells becomes impossible. The
containing different minerals of similar density. Several flotation tech­ processing of fast floating particles is therefore hydrodynamically con­
nologies have emerged, including mechanical cells, column cells and the strained by the bubble rise velocity.
Jameson cell, with alternative hybrid designs emerging in recent years. The REFLUX™ Classifier is a relatively recent density-based sepa­
There have also been significant developments in understanding the ration technology that consists of a set of parallel inclined channels
flotation chemistry and bubble-particle hydrodynamics, including de­ mounted above a fluidised bed (Galvin, 2012). Inclined channels can
velopments in flotation reagents for promoting selectivity. Despite these also be used to enhance the separation of positively buoyant particles.
advances, there continues to be a fundamental limitation in the volu­ This has led to successful research into separation of cenospheres using
metric feed flux that can be applied, directly related to the bubble rise an inverted REFLUX™ Classifier (Kiani et al., 2015). Here the inclined
velocity, which scales directly with the bubble size. Thus, the gas and channels are located below the main vertical section. This arrangement
feed flux have been constrained to about 1 cm/s. Economies of scale has forms the basis of a novel flotation technology referred to as the
led the industry to rely upon ever larger cells, with current commercial REFLUX™ Flotation Cell (RFC™), which provides significant hydrody­
cells exceeding 600 m3 in volume (Fig. 1) (Mesa & Brito-Parada, 2019). namic advantage in flotation.
This poses a significant challenge in a world where demand for the Laboratory studies have confirmed efficient operation beyond the
minerals required to enable the transition to a renewable-energy low- usual flux curve constraints, permitting feed fluxes 5–10 times higher
carbon economy is growing rapidly, whilst at the same time the than conventional flotation cells in a two-stage rougher-cleaner cell
remaining ore bodies are trending to lower grades, with the mineralogy application (Dickinson et al., 2015; Jiang et al., 2014, 2019). Recent
more finely disseminated. work has shown that the rougher-cleaner separation can be undertaken
This nominal feed flux limitation of ~1 cm/s exists because in a single stage and at throughputs up to and exceeding 5 cm/s (Cole
attempting to operate at higher feed fluxes inevitably leads to the et al., 2020; Cole et al., 2021). It is emphasized, however, that these

* Corresponding author.
E-mail address: simon.iveson@newcastle.edu.au (S.M. Iveson).

https://doi.org/10.1016/j.mineng.2022.107447
Received 8 December 2021; Accepted 8 February 2022
Available online 28 February 2022
0892-6875/© 2022 Elsevier Ltd. All rights reserved.
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

Fig. 1. Historical data on maximum commercially available mechanical flotation cell sizes. Note that the ordinate has a logarithmic scale ().
Source: Mesa & Brito-Parada, 2019

Fig. 2. Schematic of a REFLUX™ Flotation Cell with internal wash water fluidisation (Source: FLSmidth).

2
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

2. The REFLUX™ flotation cell

Fig. 2 shows a schematic of a REFLUX™ Flotation Cell. Slurry feed

enters at the top of the system via a high-shear air sparger arrangement,
and then through a downcomer into the main vertical section. Bubble-
particle collision and adhesion is generally assumed to occur prior to
downcomer discharge. The cell is enclosed at the top by a fluidisation
distribution system for delivering downwards, counter current, wash
water. The concentrate is effectively displaced upwards via a central
annulus, promoting containment of the bubbly mixture to the point of
overflow discharge. Coalescence is minimized by ensuring the gas flux,
jg, and liquid overflow flux, jl, correspond to an external gas volume
fraction jg/(jg + jl) that does not exceed ~0.90 (Sutherland et al., 2020;
Cole et al., 2021). The downwards tailings flows entrains bubbles into
the inclined channels. However, in these channels the rising bubbles
readily segregate according to the Boycott Effect (Boycott, 1920), and
return to the upper vessel, thus increasing the gas hold-up in the cell.
There is no distinct dilute bubbly zone with a froth zone above, rather
there is a single concentrated bubbly zone of gas volume fraction ~0.5
(Dickinson & Galvin, 2014).
The arrangement shown in Fig. 2 permits decoupling of the input gas
flux, jg, and liquid overflow flux, jl, delivering independent control of the
liquid flux that reports to the concentrate (Galvin & Dickinson, 2014).
The proportion of the feed flux reporting to the concentrate can be set
arbitrarily at, for example, 15 %, with no reference to maintaining the
position of an internal interface between a bubbly zone and froth zone.
The system achieves self-control of the bubble position within the in­
clined channels at an arbitrary position. This effect is powerful, and
capable of responding to significant disturbances. By ensuring the wash
water flux, jw, exceeds jl, the balance (jw – jl) results in a downwards
“positive bias flux” (Yianatos et al., 1987), which readily permeates
through the concentrated bubbly zone. The cleaning is highly efficient
Fig. 3. Photographs of the full-scale unit installed on site. because there is little or no tendency to cause coalescence, unlike with a
froth zone, and because the counter current wash water provided, jw,
supports the cleaning of a relatively high feed flux, jf.
findings have the status of mere “curiosity” until these advances can be
confirmed at full-scale. This paper reports the results from the first field
trial of a full-scale (2 m diameter) RFC™.

Fig. 4. Process flow diagram (PFD) of the full-scale research facility.

3
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

Table 1
Sample pulp density and head ash values, together with solids yield and combustible recovery calculated using Eqs. (1) and (2). Also shown are the pulp density and
head ash of the wash water sample collected at 16:00 (same time as Sample D). Values in brackets () are based on the clean wash water adjustment (see Section 3.3).
Sample Pulp Density (wt.%) Head Ash (wt. %) Solids Yield Combustible Recovery

Feed Product Reject Feed Product Reject (wt.%) (%)

A 4.5 12.5 2.8 60.3 13.8 (12.9) 83.2 33.0 (32.6) 71.6 (71.5)
B 5.5 15.0 3.4 64.2 11.3 (10.5) 80.3 23.3 (23.1) 57.8 (57.7)
C 6.1 13.9 3.5 61.3 11.1 (10.2) 81.1 28.3 (27.9) 65.0 (64.8)
D 5.2 13.4 3.0 59.9 11.6 (10.7) 80.8 30.2 (29.8) 66.6 (66.4)
E 5.2 11.3 2.8 58.8 15.5 (14.5) 83.8 36.6 (36.1) 75.1 (74.9)
Average 5.3 13.2 3.1 60.9 12.7 (11.8) 81.8 30.3 (29.9) 67.2 (67.0)
Wash Water 0.2 79.3

3. Experimental overflowing into the product sump, the reject and product pumps were
started. Then finally feed was directed into the unit with frother added.
Commercial, government and university funding enabled the Shut down of the system was done in the reverse order, with air being
installation of a 2 m diameter RFC™ unit at an operating Coal Handling the last stream to be turned off.
and Preparation Plant (CHPP) in the Hunter Valley, Australia. Details of The feed rate into the system was controlled via a feedback loop from
the machine, its operation and sampling, and how data were analysed a magnetic flow meter to the control valve. As a research system, it was
are outlined in this section. necessary to ensure both the feed and the reject flows could be
controlled, however in practice the reject flow would be adjusted in
response to fluctuations in the feed rate. Once the system was operating,
3.1. Equipment
it was allowed to approach steady state and after 30 min of approxi­
mately steady state operations, samples were collected.
The full-scale REFLUX™ Flotation Cell, which is shown in Fig. 3, had
Collecting representative samples from such large flowrates is
an internal diameter of about 2 m. The upper vertical section had a
problematic without specialised and expensive sample collection de­
height of about 1.6 m and the inclined channels were nominally 1.0 m
vices that cut the entire cross section of the stream with equal proba­
long, with a 20 mm spacing and an inclination of 70◦ to the horizontal.
bility. However, the very fine nature of the particles being sampled
As discussed, the feed slurry entered through an air sparger assembly.
means that levels of segregation should be relatively low. In this work,
The aerated bubbly feed mixture then flowed via the downcomer,
the feed and reject streams were sampled using 25 mm (1 in.) pitot tubes
entering the cell midway down the vertical section. Around the upper
facing into the flow and located at the centre of each pipe. After allowing
conical section above the main chamber were six wash water distribu­
time for the sample line to be flushed of any residual solids, the samples
tion panels through which the wash water entered.
were collected directly into 20 L buckets. The overflow bubbly product
Fig. 4 shows a simplified process flow diagram (PFD) of the system.
mixture was sampled using a splitter box to divert the entire flow for
Feed to the unit was sourced from the overflow from the classifying
several seconds into a 200 L drum lined with a plastic bag. After the froth
cyclones which had a nominal separation size of 0.100 mm. The feed
had collapsed, the slurry in this bag was then transferred to a 20 L
supply was teed off from this flotation feed line, at a point downstream
bucket. Samples of wash water were collected via a 25 mm line teed off
of where a commercial collector had already been dosed into the feed to
from the side of the wash water supply line.
independently support the plant operation at an approximate dosage of
240 g/tonne of solids. The feed pulp density was typically 5 wt% solids.
Wash water was sourced from the clarified water supply, after first
3.3. Preliminary analysis
being de-gritted in a cyclone with an internal 0.50 mm aperture screen,
followed by a back-up Y-strainer with 0.50 mm apertures. A control
All slurry samples were sent to an external accredited testing labo­
valve on the cyclone spigot opened for 5 s at 10 min intervals to
ratory for measurement of the weight percentage solids (wt.%)
discharge any accumulated solids. The wash water flowrate was
(commonly called the “pulp density”) and the head ash of the dried
adjusted using a manual valve based on values measured by a magnetic
solids as per AS 3839 and ISO 17,246 respectively. Extended Tree Curve
flow meter.
analysis was also performed on some samples as per AS4156.2.1 to
The frother used was the Nalco reagent NALFLOTE™ 9840PLUS
quantify, in principle, the maximum cumulative flotation yield as a
dosed by air-actuated pumps from an intermediate bulk container (IBC).
function of the cumulative ash %.
This frother was first diluted via a pair of in-line mixers in order to form
Other samples were sent to a different external laboratory for coal
emulsions which were then pumped to two different locations, namely:
grain analysis (CGA). CGA is a relatively new optical technique devel­
(i) the feed slurry immediately after it was teed off from the plant; and
oped by CSIRO (Australia) that uses reflectance measurements of sam­
(ii) the wash water after it exited the de-gritting cyclone. A nominal
ples set in a polished resin block to determine the distribution of particle
frother dosage of 20 ppm was used in both the feed and wash water.
size and grain types. From this data together with assumptions about
The underflow reject stream was pumped by a variable speed pump
coal maceral densities and ash contents, the cumulative yield versus
to the thickener. The flow was kept at a constant rate by a feedback loop
cumulative ash curve can be predicted (www.csiro.au). This method
from a magnetic flow meter. The overflow rate was not controlled
potentially provides an alternative to the sink-float method which
directly, but rather was determined by the difference between the
cannot be easily applied to fine particles.
combined input feed and wash water, less the rate of underflow removal.
It is often the case that the inferred head ash values from the CGA
The overflow bubbly mixture product was collected in a deaeration
analyses are lower than the measured head ashes of the samples. This
sump from where it was then pumped back to the plant.
error arises due to the 1 μm limit of resolution, and the fact that ultrafine
mineral matter, including clays, is often significantly finer. Other errors
3.2. Operation arise due to poor colour contrast. In the present study, the differences
were larger than usual, with the CGA inferred ashes in the range 40–45
Operation of the system commenced with setting the air flow and wt% ash, whereas the actual head ashes were in the range 59–64 wt%.
then the wash water addition. Once the unit was full of water and To correct for this error, it is standard practice to add a quantity of high

4
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

Table 2
Average sizes, corrected ashes and calculated yields and combustible recoveries from the oil-agglomeration separation of hydrophobic and hydrophilic portions of the
Sample E feed, product and reject samples. See Appendix A1 for details of the calculations.
Sample E Feed Product Reject

Total Hydro- Hydro- Total Hydro- Hydro- Total Hydro- Hydro-

Phobic Philic Phobic Philic Phobic Philic

Head Ash, A (wt.%): 58.8 15.5 83.8

Yield to Product (Eq. 1 using head ashes) (wt.%): 36.6
Combustible Recovery (Eq. 2 using head ashes) (%): 75.1
Laser Sizer Data:
Sauter mean, d32 (µm): 4.1 16.2 2.6 12.6 21.1 4.7 2.5 11.7 2.4
Volume mean, d43 (µm): 36.6 58.3 13.3 63.0 64.4 31.5 12.4 37.5 9.2
Oil Agglomeration Method:
Raw Ash, AH or AL (wt.%): 10.6 84.9 7.7 66.6 20.9 87.0
Corrected Ash, AHC or ALC (Eq. 4 using y ¼ 0.8) (wt. 11.5 85.8 8.0 70.2 25.5 87.3
%):
Hydrophobic Yield (Eq. 1 using corrected ashes) (wt. 36.4 88.0 5.6
%):
Combustible Recovery (Eq. 2 using corrected ashes) 78.0 95.8 26.0
(%):

ash material to the raw CGA data to bring the inferred ash into line with slimes. This approach also provides a measure of the potential product
the measured head ash (Cole et al., 2020). We chose to add material with quality, and that of the unrecovered material that reported to the reject.
87 wt% ash, as this was the highest ash of the hydrophilic fraction A new method of analysis was applied in order to examine the hy­
recovered during the analysis (Table 2). Appendix A3 shows both the drophobic particles in the feed, product, and reject streams. The novel
raw and corrected CGA curves for comparison. method of oil agglomeration described by van Netten et al. (2017) ex­
From the head ashes of each flow stream, the mass fraction yield Y of tracts particles that exhibit almost any level of hydrophobicity, across
feed solids F to the product P and the combustible recovery R were the full particle size range, in just a few seconds. The method uses a
calculated using the two-product formula (Wills & Finch, 2016): water-in-oil emulsion as the binder, consisting of 95 wt% water, 2.5 wt%
kerosene and 2.5 wt% emulsifier. The emulsifier was an industrial grade
P (xR − xF )
Y= = (1) sorbitan monooleate (SMO, C24H44O6) mixture consisting of 60 wt%
F (xR − xP )
SMO and 40 wt% glycerol monooleate. It is noted that in the current
work no salt was included in the aqueous phase. A known amount
(100 − xP )P (100 − xP )(xR − xF )
R= = (2) (15–20 g) of this binder was added to a known mass of slurry (150–300
(100 − xF )F (100 − xF )(xR − xP )
g) and then sheared at 22 000 rpm in a 1 L Waring blender (Model
where xF, xP and xR are the head ashes of the feed, overflow product LB20EG) for 7 s. Hydrophobic particles were captured in the oil ag­
and underflow reject respectively. These formulae assume the process glomerates which were then recovered by pouring the blended slurry
was at steady state, the samples were representative, that there was no over a 0.25 mm aperture screen. The hydrophilic particles passed
sample reaction or degradation inside the process or during storage, and through the screen, assisted by light levels of wash water. Solids from
that the mass of ash is equal to the mass of mineral matter in the coal both fractions were recovered by drying overnight in an oven at
samples. 105–110 ◦ C. Laser particle size analysis was applied to the solids from
Implicit in using these equations is the assumption that there were no each of the flow streams, and other samples were sent to an external
solids present in the wash water. However, even after passing through laboratory for head ash analysis. These ash values were then corrected to
the de-gritting cyclone and Y-strainer, the wash water supply did still account for the presence of non-volatile SMO in the samples (see Ap­
contain a significant loading (0.2 wt%) of high-ash (79.3 wt%) slimes pendix A1).
(Table 1). Ideally, using clean water, the wash water displaces all the Particle size distributions of feed, product and reject slurry samples
entrained slimes in the overflow product, leaving only the hydrophobic were measured using a Malvern Mastersizer 3000. The size distributions
material that was attached to air bubbles and clean water. We have of the dried hydrophobic and hydrophilic fractions separated by the oil-
estimated the potential improvement that might have been obtained agglomeration method were also measured. These data, combined with
using clean wash water by calculating the effect of removing from the the mass splits of the material, enabled the recovery to the product of
overflow product a total mass and ash of solids equivalent to that carried hydrophobic material to be calculated as a function of particle size (see
in the wash water. For example, product Sample A had a pulp density of Appendix A2). Mass yields were converted to volume splits using the
12.5 wt% with 13.8 wt% ash (Table 1). So 100 g of slurry contained 12.5 solid density ρ of each sample. This density was estimated from the head
g of solids containing 0.138(12.5 g) = 1.725 g of ash, and 87.5 g of water ash A (wt.%) by assuming the particles were a binary mixture of
which (assuming it was identical to the wash water) contained 0.002 combustible material of density ρcomb = 1.2 g/mL and non-combustible
(87.5 g) = 0.175 g solids containing 0.793(0.175 g) = 0.139 g of ash. So (ash) material of density ρash = 2.6 g/mL using a corrected form of the
replacing this dirty wash water with clean water would give 12.5 g – equation given in Luttrell et al. (2003):
0.175 g = 12.325 g of product solids containing 1.725 g – 0.139 g =
100
1.586 g of ash i.e. 12.9 wt% ash, so roughly a 1 wt% ash reduction. ρsolid = A
(3)
ρash + 100−
ρ
A
comb

3.4. Analysis of hydrophobic particles It is noted that a sensitivity analysis varying in turn the assumed ρash
and ρcomb values by ± 10 %, produced variations of less than 0.2 vol% in
There is interest in examining the hydrophobic particles captured in the calculated volume splits of each component between the product
the flotation product concentrate, and those hydrophobic particles that and reject.
fail to be captured, and hence report to the reject stream. Such analysis
helps to quantify precisely the recovery of the particles as a function of
the particle size, and the extent to which the system rejected ultrafine

5
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

thus no additional collector was added prior to the lab-scale test. Methyl-
isobutyl-carbinol (MIBC) frother was, however, added at 20 ppm to the
feed to ensure a stable froth was formed during flotation. The impeller
speed was set at 700 rpm and the air rate at 10 L/min, equating to a gas
flux of 0.8 cm/s. As froth was removed from the cell, additional water
containing 20 ppm MIBC frother was added to maintain the slurry level.
The initial test ran for 6 min. The tailings was set aside and the product
was then re-suspended and re-floated for another 6 min, however it was
noted the froth was virtually barren of solids after ~ 4 min. Then the
secondary product was re-floated again for a third time for 6 min (clear
froth was achieved after ~ 3 min). The four sub-samples (Tailings 1,
Tailings 2, Tailings 3 and Product 3) were dried, weighed and sent to an
external laboratory for percent ash measurement. From these data a
cumulative yield versus cumulative ash curve with four points could be
obtained. It is noted that municipal water with 20 ppm MIBC frother was
used to re-suspend the product obtained at each step before it was re-
floated.
Fig. 5. Feed, product and reject head ashes versus time. Lines show the
average values.
4. Results and discussion

4.1. System stability and separation performance

This paper reports the results of a series of full-scale REFLUX™

Flotation Cell experiments conducted over a period of four hours of
nominally steady set conditions. This was the first occasion on which the
fully-commissioned machine had been able to run for such a length of
time with a stable feed from a single feed source being supplied by the
plant. Results are bench-marked against tree flotation curves,
laboratory-scale mechanical cell multistage cleaner tests, and coal grain
analysis. This paper also includes a detailed investigation of one set of
samples using a novel oil-agglomeration analysis technique to measure
the hydrophobic and hydrophilic portions, to obtain the recovery versus
size of these two fractions.
Five sets of samples (labelled A, B, C, D and E) were collected during
the afternoon (at approximately 13:25, 13:55, 14:25, 16:00 and 16:30).
Fig. 6. Combustible recovery (from Eq. (2)) versus reject ash for a product ash Frother addition was started at 12:50. Over the next 20 min the pressure
of 11 wt% and three different feed ashes. Note the difficulty in obtaining high at the base of the unit decreased from its intial value of 40 kPa when full
combustible recovery when the feed ash is high. of only water, eventually reaching a value of around 30 kPa. This
lowering in pressure indicates that bubbles were accumulating in the
vertical section. A 10 kPa change is consistent with the top 2 m of the
unit reaching a bubble volume fraction of 0.5 as observed in laboratory-
scale systems, consistent with the formation of a concentrated bubbly
zone beyond the point of flooding.
The air flowrate was kept constant at the set point of 180 m3/h. The
feed rate was maintained mostly in the range 200 to 220 m3/h, equiv­
alent to a volumetric feed flux of ~ 1.9 cm/s, about twice the normal
level in conventional open flotation cell designs. Variations in the feed
rate were due to changes in the plant operation, and also vagaries of the
feed control loop which had not yet been properly tuned. The feed pulp
density varied in the range 4.5 to 6.1 wt%. The wash water rate was kept
at about 65 m3/h throughout the day and the underflow rate at 235 m3/
h. Based on measurements from the control system, it was concluded
that the system was inherently stable prior to and during the sampling
periods, despite there being some early teething issues with the
controller. The calculated overflow varied from 35 to 55 m3/h (due to
the feed rate variations), giving downwards positive wash water bias
flows of 10 to 30 m3/h, equivalent to bias fluxes of 0.09 to 0.27 cm/s.
The pulp density and head ash of the five sets of feed, product and
Fig. 7. Product ash versus pulp density. Line included to guide the eye. reject samples were measured and the results are summarised in Table 1.
These had an average feed ash of 60.9 wt%, product ash of 12.7 wt%,
3.5. Mechanical cell bench marking and reject ash of 81.8 wt% across the period. Fig. 5 shows the head ash
values versus time, demonstrating the very stable performance of the
The full-scale separations were also compared with flotation results system during the trial. The variations in the product and reject ashes
from a laboratory-scale 3.5 L mechanical flotation cell (FLSmidth Essa were of the same order of magnitude as the variation in the ash of the
model FTM101). The feed for the mechanical cell tests had previously feed into the system, which was a function of the industrial variations at
been conditioned with collector on-site (prior to sample collection), and the plant. This degree of steady operation over this period represents the

6
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

Fig. 8. Cumulative yield versus ash showing results from the full-scale unit (Samples A-E raw and with clean wash water adjustment), the tree curve analyses and
adjusted CGA analyses, and separation of Sample E in a laboratory-scale mechanical cell with long residence times. Sample collection times shown in top-left corner.

first confirmation that the system hydrodynamics is scalable. in Appendix A3.

It is also worth noting that the feed ash wt.% levels involved here are Fig. 8 shows the cumulative yield versus cumulative ash results.
relatively high at an average of 60.9 wt% with a range from 58.8 to 64.2 There are three sets of extended tree curve data sets, taken from samples
wt%. Thus, the yield values which average 30.3 wt% are relatively low, collected at different times in the afternoon. The two labelled “IBC” refer
reflecting the low content of coal in the feed. Moreover, it is emphasized to feed samples collected into an intermediate bulk container (1000 L)
that for coal separations, a reject ash of 80 wt% or more is deemed very over a period of about 20 min (collection times indicated in the Fig. 8
high by Australian standards. In particular, the combustible recovery is top-left corner). The Sample E full-scale RFC™ result lies just on or
very sensitive to the reject ash and feed ash. Fig. 6 shows the theoretical slightly above the extended tree curve analysis of Sample E feed. The
sensitivity of the combustible recovery to changes in the reject ash for other four full-scale results (Samples A-D) all lie to the left of and above
three hypothetical feeds of varying ash level at a fixed product ash. High the tree curves of samples IBC#2 and IBC#3, which cover the periods
combustible recoveries are readily achieved at low feed ashes, but for when samples C and D were collected. Hence in every case, the full-scale
the high feed ash of 60 wt% (which is indicative of the feed in the RFC™ has performance that is equal to or better than the tree curve of
present study), very high reject ashes are required to obtain high re­ the feed at that point of time. If the potential improvement from using
coveries. The highest ash hydrophilic material obtained in the lab in this clean wash water is considered (also shown on plot), then the full-scale
study was 87 wt% ash (Table 2), which would only just give 80 % unit has out-performed the tree curve in every case. This performance is
combustible recovery if it could be attained during actual operation. even more remarkable given the fact that the feed is exceedingly fine,
Thus, the average combustible recovery of 67 % achieved here should be the feed ash is very high, and the feed flux is 1.9 cm/s, well in excess of
interpreted with due consideration to the difficultly of the separation. typical flotation feed rates.
Fig. 7 shows a strong correlation between the product ash and the Also shown in Fig. 8 are the corrected coal grain analysis (CGA) re­
product pulp density. The increase in product ash at lower pulp densities sults (Section 3.3). There is reasonable agreement between these and the
reflects the increased levels of hydrophilic (high ash) material in the tree curve results, although the CGA analysis predicts unrealistically low
entrained water. Hence it is important to have good control of the inherent ashes below 1 wt%, and also above the “knee” of the curve the
underflow rate in response to variations in feed rate to maintain a CGA results are often below the tree curve results. In this work the CGA
constant water split to the overflow. Note that the wash water being results are relatively close to those of the tree flotation analysis, which
used had 0.2 wt% solids with 79 % ash (Table 1), so using cleaner wash suggests the tree flotation analysis is actually a good measure in this case
water and/or dewatering the product should reduce the product ashes of strong flotation performance. However, it is noted that the tree curve
by up to 1 wt% (see bracketed values in Table 1). can be unreliable, which is why there has been so much interest in the
CGA method, as this generally indicates the true limits of cleaning.
Either way, it can therefore be concluded that the RFC™ has achieved an
4.2. Flotation yield versus ash
extreme level of cleaning at full-scale.
Fig. 8 also shows the performance of the laboratory-scale mechanical
Samples of the feed were investigated to quantify the maximum
cell on Sample E. The mechanical cell results reside slightly above the
possible yield as a function of the cumulative ash %. Tree curve flotation
tree curve and CGA results, which is unsurprising given the batch nature
is an industry standard method used to give some indication of the best
of the system, multistage operation, and use of long residence times and
possible flotation performance. Coal grain analysis (CGA) is a more
clean wash water. Moreover, the first stage mechanical stage separation
recently developed optical technique that has also proven effective in
produced a product ash of 22 wt%, which was lowered at each stage.
quantifying flotation performance. For this feed it was necessary to
Again, it is remarkable that the continuous full-scale RFC™ operated at
apply a significant correction to the CGA data due to the high portion of
such high feed flux was able to achieve performance so close to the best
particles finer than 1 μm, the limit of resolution. This correction is shown

7
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

Fig. 9. Size distributions of the Sample E feed, product and reject samples as measured by Malvern Mastersizer 3000. a) Raw data showing volume fraction in
interval versus interval size on a logarithmic scale and b) plot of volume frequency (volume percentage divided by interval width) versus interval size on a linear
scale truncated at 30 μm.

that could be obtained in a laboratory under ideal conditions with long the nominal cut size of the cyclone supplying the flotation feed. The
residence times. It is worth noting the product flux from the continuous three curves intersect at around 20 μm, indicating some degree of mass-
full-scale RFC™ was ~ 57 times higher than for the 6 min batch test. balance consistency. However, the variable interval widths and loga­
rithmic scale obscure the true character of the size distribution. Fig. 9b
4.3. Hydrophobic-Hydrophilic separation of samples shows the volume percentage divided by the interval width (i.e. the
volume frequency) plotted versus interval average size on a linear scale
The size distributions of the Sample E feed, product and reject, which has been truncated at an upper size of 30 μm in order to make the
measured using a Malvern Mastersizer 3000, are shown in Fig. 9. Fig. 9a details at finer sizes clearer. In this plot, the area under the curve is
shows the raw data from the Malvern on a logarithmic size scale which indicative of where the material is distributed. This plot clearly dem­
makes the details clear and shows the drop off above 100 μm, which was onstrates the extremely fine nature of the feed. The volume mean sizes

8
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

Fig. 10. Size distributions of head, hydrophobic and hydrophilic portions of Sample E a) feed, b) product and c) reject. Head sample raw ash and corrected ashes of
hydrophobic and hydrophilic portions also indicated.

were 36.6, 63.0 and 12.4 μm and the Sauter mean sizes were 4.1, 12.6 Sample E feed, product and reject were separated using the oil
and 2.5 μm for the feed, product and reject respectively. This is an agglomeration technique of van Netten et al. (2017) as described in
extremely fine feed with 50 vol% of material less than 14.4 μm in size. Section 3.4. This technique has been demonstrated to be a very efficient
The hydrophobic and hydrophilic fractions of sub-samples from the and rapid method for separating the hydrophilic and hydrophobic

9
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

these calculation methods, please refer to Section 3.4 and the appendix.
If we first focus on the oil-agglomeration method itself, it can be seen
that it was very effective at capturing low-ash (hydrophobic) particles.
For instance, the feed sample with 58.8 wt% head ash was split into a
hydrophobic portion with corrected ash of 11.5 wt% ash and a hydro­
philic portion with a corrected ash of 85.8 wt% (Table 2). The corrected
ashes of the hydrophobic fractions were 11.5, 8.0 and 25.5 wt%
respectively for the feed, product and reject samples, and the corrected
ashes of the hydrophilic fractions were 85.8, 70.2 and 87.3 wt%
respectively (Table 2). These numbers compare favourably with the
tree-curve results for Sample E, with the lowest and highest ashes ob­
tained being 8.2 wt% and 87.4 wt% respectively. The relatively high
25.5 wt% ash of hydrophobic material in the reject sample and relatively
low 70.2 wt% ash of hydrophilic material in the product sample both

Fig. 11. Sample E full-scale RFC™ hydrophobic and hydrophilic recovery

versus particle size, calculated from the laser sizing of the oil agglomeration
analysis data in the 1–50 μm size range.

portions of a sample, even down to very fine (below 1 μm) particle sizes.
Hence it provides a means for quantifying separation performance
versus size for materials that are too fine to assess by the standard sink-
float method, such as many flotation feeds. This paper is the first pub­
lication we are aware of where the oil-agglomeration technique is used
for this purpose.
Table 2 shows raw data from the oil-agglomeration separations and
laser sizings, together with various corrections made to account for the
presence of non-volatile SMO in the dried samples. From the corrected
ashes, various yields and recoveries are calculated on a basis of 100 g
total of feed. These mass data are then converted to volume data which
is combined with the laser sizing results to calculate the recovery of Fig. A1. Raw and adjusted cumulative yield versus cumulative ash curves from
hydrophobic material as a function of particle size. For more details on the CGA analysis of IBC#2, IBC#3 and Sample E.

Table A1
Raw data from oil-agglomeration separations of Sample E feed, product and reject together with details of the SMO ash correction and mass to volume conversion
calculations.
Sample E Feed Product Reject

Total Hydro- Hydro- Total Hydro- Hydro- Total Hydro- Hydro-

Phobic Philic Phobic Philic Phobic Philic

Head Ash, A (wt.%): 58.8 15.5 83.8

Yield to Product (Eq. 1 using head ashes) (wt.%): 36.6
Combustible Recovery (Eq. 2 using head ashes) (%): 75.1
Oil Agglomeration Method:
Raw Ash, AH or AL (wt.%): 10.6 84.9 7.7 66.6 20.9 87.0
Raw Mass, mH or mL (g): 5.1 10.0 10.3 2.0 1.7 22.8
Emulsion used, mB (g): 20.8 20.7 15.2
SMO in Emulsion (g): 0.5 0.5 0.4
Assumed SMO distribution (y = 0.8) (g): 0.4 0.1 0.4 0.1 0.3 0.1
Corrected Mass (i.e. without SMO), mHC or mLC (g): 4.7 9.9 9.9 1.9 1.4 22.7
Hydrophobic Yield (from corrected masses) (wt.%): 32.1 83.9 5.8
Corrected Ash, AHC or ALC (Eq. 4 using y ¼ 0.8) 11.5 85.8 8.0 70.2 25.5 87.3
(wt.%):
Ash Correction Error Range (y = 0.6 to y = 1) (wt. 11.3–11.8 84.9–86.7 7.9–8.1 66.6–74.3 24.1–26.9 87.0–87.6
%):
Hydrophobic Yield (Eq. 1 using corrected ashes) 36.4 88.0 5.6
(wt.%):
Combustible Recovery (Eq. 2 using corrected ashes) 78.0 95.8 26.0
(%):
Mass to Volume Conversions for Calculating Recovery by
Size
Density (Eq. 3 using head or corrected ashes) (g/ 1.756 1.280 2.230 1.309 1.254 1.930 2.187 1.391 2.264
mL):
Mass distribution (based on F = 100 g and Yields) 100.0 36.4 63.6 36.6 32.2 4.4 63.4 3.6 59.8
(g):
Volume distribution (dividing mass by density) 56.9 28.4 28.5 28.0 25.7 2.3 29.0 2.6 26.4
(mL):
Check Volume balances In/Out ratios: 1.000 1.006 0.994
Volume Yield to Product (vol.%): 49.1 90.4 8.0

10
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

reflect the likely presence of significant numbers of composite (poorly The coal feed slurry head ash varied in the range 59–64 wt% head
liberated) particles in those two sample fractions. The oil agglomeration ash. From this product ashes of 11–15 wt% were obtained at combus­
method captured 95.8 % of the combustible material in the product tible recoveries of 58–75 %, with results on or better than the flotation
sample (yield of 88 wt%), 78.0 % of the combustibles in the feed (yield tree curve and CGA data, and only slightly below results obtained from a
of 36.4 wt%), and 26.0 % of the combustibles in the reject (yield of 5.6 bench-scale mechanical cell operated in batch mode with multiple
wt%). Again, the high starting ash of the reject sample made it hard to stages, long residence times, and clean wash water. Across the size range
achieve high combustible recoveries at a low ash grade (recall Fig. 6). 1–50 μm, the recovery of the hydrophobic particles was 83 % or higher
Fig. 10 a-c shows the size distributions of the Sample E feed, product and recovery of hydrophilic particles was less than 20 %, becoming less
and reject, together with the size distributions of the hydrophobic and than 10 % at particle sizes below 10 μm. These results demonstrate that
hydrophilic portions obtained using the oil-agglomeration analysis the favourable hydrodynamics obtained in laboratory-scale units extend
method. The plots show the raw Malvern Mastersizer 3000 data volume to full-scale machines. The feed flux used was roughly 1.9 cm/s, which is
percentage in each interval versus interval average size on a logarithmic about twice the rate that conventional cells are usually operated at, a
scale (recall that as shown in Fig. 8 these plots distort the size distri­ significant step-change in capacity.
bution). The Sauter and volume mean sizes of all portions are reported in
Table 2. The hydrophobic fractions are all much coarser than the cor­
Declaration of Competing Interest
responding hydrophilic ones, as shown clearly both in the size distri­
butions and the average sizes. It is also clear that the product is nearly
Aside from the disclaimer below, the authors declare that they have
entirely hydrophobic material, and the reject is nearly entirely hydro­
no other known competing financial interests or personal relationships
philic material.
that could have appeared to influence the work reported in this paper.
Fig. 11 shows full-scale RFC™ hydrophobic and hydrophilic recov­
ery versus particle size, calculated based on the laser sizing and oil
agglomeration data (see Appendix A.1 and A.2 for details of these cal­ Acknowledgments and Disclaimers
culations). Note that results are only shown for the 1–50 μm size range
due to the poor mass balance consistency outside this range, likely noise The University of Newcastle holds international patents on the
due to the small amount of material present at finer and coarser sizes. REFLUX™ Flotation Cell and has a Research and Development Agree­
The recovery of hydrophobic material is 83 % or higher across these ment with FLSmidth Pty Ltd. One of the report authors, K. P. Galvin, is
sizes, increasing to above 90 % at the coarser sizes. In contrast, the re­ the inventor of the REFLUX™ Flotation Cell and is a beneficiary of the
covery of hydrophilic material is less than 20 % across this range, and University’s intellectual property policy.
remarkably less than 10 % below 10 μm in size, giving selectivity ratios This research was funded by an Australian Government Global
approaching 20 or more. The slimes rejection would have been even Innovation Linkage grant (GIL53761) and the Australian Coal Associa­
better if cleaner wash water had been available. tion Research Program (ACARP Project C23045). The upscaling of the
technology also is aligned with the research programs of the ARC Centre
5. Conclusions of Excellence for Enabling Eco-Efficient Beneficiation of Minerals (Grant
CE200100009). A number of companies also contributed valuable sup­
A full-scale 2 m diameter REFLUX™ Flotation Cell (RFC™) has been port to this project:
successfully commissioned and operated at a coal preparation plant for
the first time, following nearly ten years of research. This paper pre­ • FLSmidth fabricated and installed the REFLUX™ Flotation Cell and
sented a detailed investigation of the unit operated steadily for a 4 h provided commissioning and maintenance support.
period. The data showed variations in product and reject ashes of a • We thank the site for access to their flotation feed and electrical
similar magnitude to the variations in the ash of the feed supplied by the supply. We particularly thank the site managers, engineers, opera­
plant. A feed rate of 200–220 m3/h was processed, containing 5 wt% tors, electricians and other site staff and contractors who patiently
solids with a very fine volume median diameter of 0.014 mm. Fixed air, gave their time to help us install, modify and operate the facility.
wash water and underflow rates were used of 180 m3/h, 65 m3/h and • Nalco provided the frother and the reagent dosing system used in this
235 m3/h respectively, giving a positive downwards wash water bias work.
flux of about 0.2 cm/s. • Advitech provided in-kind support during the design of the facility.

Appendix

Appendix A:. Details of additional data analysis

A.1 Correction for non-volatile SMO in oil agglomeration ash analysis

SMO is a relatively non-volatile liquid (boiling point of order 500 ◦ C), and so it is not removed during drying but is removed during the combustion
process used to measure ash content. Hence its presence causes the measured head ashes to be too low. Let mH and mL be the dried mass of hydrophobic
and hydrophilic material respectively in a sample, and let AH and AL be the raw head ash measured for each sample (“H” for water “hating” and “L” for
water “loving”). If the mass of binder used was mB, then the mass of SMO was 0.025mB. If we let y be the fraction of this SMO that reports with the
hydrophobic solids, then the corrected (C) estimate of the true head ash of each fraction was calculated by:
AH mH AH m H
AHC = = (4a)
mHC mH − 0.025mB y

AL mL AL m L
ALC = = (4b)
mLC mL − 0.025mB (1 − y)
The value of y is unknown. Based on the work of Borrow et al. (2018), it is known that around 40 % of the water in the binder emulsion is released
during blending, so potentially up to 40 % of the SMO might report with the hydrophilic portion, leaving y = 60 % with the hydrophobic portion.

11
S.M. Iveson et al. Minerals Engineering 179 (2022) 107447

However, the SMO is active at the water-kerosene interface, and so arguably it remains with the kerosene which will largely report with the hy­
drophobic particles (i.e. y = 100 %). Qualitatively it was observed that the dried hydrophilic fraction of the product sample was “sticky”, which is not
the case in normal dried coal samples, and this is attributed to the presence of residual liquid SMO. So clearly some SMO did still report with the
hydrophilic fractions. Therefore an assumption of y = 0.8 seems reasonable and is used to obtain the corrected ash values shown in Table 2 and
Table A1. Table A1 also shows the range of corrected ash values obtained using y = 0.6 and y = 1.0, to indicate the sensitivity of the results to this
assumption. In most cases the variation in the estimated true ash is less than ± 1 wt% ash. The true ash of the hydrophilic portion of the product has the
greatest uncertainty, due to it being one of the smallest subsamples and also having the least amount of combustible material to begin with. Its
uncertainty range is from 66.6 to 74.3 wt% ash, so ±4 wt% ash.

A.2 Calculation of hydrophobic and hydrophilic recovery by size

Using the head ash and the corrected ash values for the hydrophobic and hydrophilic fractions, Eq. (1) was used to calculate the mass fraction of the
solids in the hydrophobic and hydrophilic fractions of each stream. Table A1 shows that these yields (36.4, 88.0 and 5.6 wt% for the feed, product and
reject respectively) are in fairly close agreement with the yields calculated directly from the corrected mass of hydrophobic and hydrophilic solids
(32.1, 83.9 and 5.8 wt% respectively). The corrected ashes were also used in Eq. (2) to calculate the recovery of combustible material in the hy­
drophobic fraction from each stream, giving values of 78.0, 95.8 and 26.0 % respectively for the feed, product and reject (Table A1).
Assuming a basis of 100 g of feed, the Eq. (1) yield based on head ashes gave 36.6 g of product and 63.4 g of reject. Then the hydrophobic yields
based on corrected ashes were used to calculate the masses of hydrophobic and hydrophilic material in all three streams. Each of these masses was then
divided by the density found using corrected ash in Eq. (3) to determine the volume of each fraction of material. Inferring particle density from ash
content by using Eq. (3) conserves the mass and volume balance consistency, but the assumption of a binary mixture of only two different pure solids is
problematic. However, there was excellent internal consistency of the adjusted ash data, demonstrated by the fact that the volume ratios of feed
divided by combined product and reject were 1.006 and 0.994 for the hydrophobic and hydrophilic portions respectively (Table 2). These volumes
were then used to determine that yields of 49.1 vol% of the total feed, 90.4 vol% of the hydrophobic material and 8.0 vol% of the hydrophilic material
reported to the product stream. These volume splits combined with the laser size volume distributions then enabled the recovery as a function of size to
be calculated (Fig. 11).

A.3 Coal grain analysis adjustment

Fig. A1 shows that the apparent head ashes of 40–50 wt% obtained from the raw CGA analysis were much lower than the measured head ashes of
the samples (around 60 wt% ash). As discussed, this is not unusual, and is attributed to there being a significant amount of high ash slimes with particle
size less than 1 μm, which were too small to be resolved optically. To correct for this, a quantity of high ash (87 wt%) material was added to the data, so
that the CGA head ash would then match the directly measured head ash for that sample. The raw and adjusted CGA yield versus cumulative ash curves
are shown in Fig. A1 for comparison.

References Jiang, K., Dickinson, J.E., Galvin, K.P., 2019. The kinetics of Fast Flotation using the
Reflux Flotation Cell. Chemical Engineering Science 196, 463–477. https://doi.org/
10.1016/j.ces.2018.11.012.
Borrow, D.J., van Netten, K., Galvin, K.P., 2018. Ultrafine Particle Recovery Using Thin
Kiani, A., Zhou, J., Galvin, K.P., 2015. A Pilot Scale Study of Cenosphere Recovery and
Permeable Films. Frontiers. Chemistry 6. https://doi.org/10.3389/
Concentration using the Inverted Reflux Classifier. Minerals Engineering 79, 17–23.
fchem.2018.00220. Article 220.
Luttrell, G.H., Barbee, C.J., Stanley, F.L., 2003. Optimum cut points for heavy medium
Boycott, A.E., 1920. Sedimentation of blood corpuscles. Nature 104 (2621), 532.
separations. In: Honaker, R.Q., Forest, W.R. (Eds.), Advances in gravity
Cole, M.J., Dickinson, J.E., Galvin, K.P., 2020. Recovery and cleaning of fine
concentration. Society for Mining, Metallurgy and Exploration, pp. 81–91.
hydrophobic particles using the Reflux™ Flotation Cell. Separation and Purification
Mesa, D., Brito-Parada, P.R., 2019. Scale-up in froth flotation: A state-of-the-art review.
Technology 240, 116641. https://doi.org/10.1016/j.seppur.2020.116641.
Separation and Purification Technology 210, 950–962. https://doi.org/10.1016/j.
Cole, M.J., Dickinson, J.E., Galvin, K.P., 2021. Maximising recovery, grade and
seppur.2018.08.076.
throughput in a single stage Reflux Flotation Cell. Minerals Engineering 163. https://
Sutherland, J.L., Dickinson, J.E., Galvin, K.P., 2020. Flotation of coarse coal particles in
doi.org/10.1016/j.mineng.2020.106761.
the Reflux™ Flotation Cell. Minerals Engineering 149, 106224. https://doi.org/
Dickinson, J.E., Galvin, K.P., 2014. Fluidized bed desliming in fine particle flotation –
10.1016/j.mineng.2020.106224.
Part I. Chemical Engineering Science 108, 283–298. https://doi.org/10.1016/j.
van Netten, K., Borrow, D.J., Galvin, K.P., 2017. Fast Agglomeration of Ultrafine
ces.2013.11.006.
Hydrophobic Particels Using high-Internal-Phase Emulsion Binder Comprising
Dickinson, J.E., Jiang, K., Galvin, K.P., 2015. Fast flotation of coal at low pulp density
Permeable Hydrophobic Films. Industrial & Engineering Chemistry Research 56
using the Reflux Flotation Cell. Chemical Engineering Research and Design 101,
(38), 10658–10666. https://doi.org/10.1021/acs.iecr.7b02215.
74–81. https://doi.org/10.1016/j.cherd.2015.04.006.
Wills, B.A., Finch, J.A., 2016. Will’s Mineral Processing Technology: An Introduction to
Galvin, K.P., 2012. Development of the Reflux Classifier. In: Kilma, M.S., Arnold, B.J.,
the Practical Aspects of Ore Treatment and Mineral Recovery, (8th ed.). Butterworth-
Bethell, P.J. (Eds.), Challenges in Fine Coal Processing, Dewatering and Disposal.
Heinemann (Elsevier).
SME.
Yianatos, J.B., Finch, J.A., Laplante, A.R., 1987. Cleaning action in column flotation
Galvin, K.P., Dickinson, J.E., 2014. Fluidized bed desliming in fine particle flotation –
froths. Transactions of the Institute of Minerals and Metallurgy C-Mineral Processing
Part II: Flotation of a Model Feed. Chemical Engineering Science 108, 299–309.
and Extractive Metallurgy 96, C199–C205.
https://doi.org/10.1016/j.ces.2013.11.027.
Jiang, K., Dickinson, J.E., Galvin, K.P., 2014. Maximizing bubble segregation at high
liquid fluxes. Advanced Powder Technology 25 (4), 1205–1211. https://doi.org/
10.1016/j.apt.2014.06.003.

12