Minerals Engineering 109 (2017) 126–134

Contents lists available at ScienceDirect

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

Slurry rheology prediction based on hyperspectral characterization models MARK

for minerals quantification
⁎
Javier Merrilla, , Leandro Voisina,c, Victor Montenegrob, Christian F. Ihlea,c, Angus McFarlaned
a
Advanced Mining Technology Center (AMTC), Universidad de Chile, Tupper 2007, Santiago, Chile
b
CSIRO Chile International Centre of Excellence, Apoquindo 2827, 12 Floor, Las Condes, Santiago, Chile
c
Department of Mining Engineering, Universidad de Chile, Tupper 2069, Santiago, Chile
d
CSIRO Mineral Resources Australia, 26 Dick Perry Av., Kensington, Australia

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

Keywords: The presence of clays in mineral processing offers a number of challenges that range from valuable species
Rheology recovery to the transport of tailings. In particular, when the abundance of one or more clay types increases, the
Suspension rheology may be significantly affected. In this paper, the feasibility of using hyperspectral characterization to
Spectroscopy estimate rheological properties of mineral suspensions was studied. Towards this purpose, a set of rheology
Clay
measurements was made for slurries of different composition, combining up to three out of five minerals: three
Mineral processing
clay minerals (two bentonites from different sources and kaolin), quartz and white mica, which are the main
Geometallurgy
gangue minerals present in the Chilean copper mining industry. Using a Bingham Plastic flow model, a set of
ternary plots for Bingham viscosity and yield stress was obtained. Results show counter-intuitive behavior for
kaolin-white mica mixtures, showing a minimum for viscosity at a 2:3 ratio respectively. In addition,
mechanisms for lowering the high viscosity reached by bentonite slurries were assessed. Modelling of the
hyperspectral data produced high accuracy estimates of the mineral abundances, enabling an accurate
determination of the respective samples position in the ternary mineralogy-rheology diagrams.

1. Introduction rheological problems due to their common occurrence in metalliferous

The rheological behavior of slurries is very important in metallur-
gical processing plants, as it can act as a limiting factor in treatment
capacity, due to the fact that at high viscosity and/or yield stress pumps
and other equipment may malfunction (Klein and Hallbom, 2002; Kelm
and Helle, 2005; Wang et al., 2015; Cruz et al., 2013; Zhang and Peng,
2014). Some mineral mixtures may show unexpected rheological
properties (Tan et al., 2013, 2012), such as; sharp rises in viscosity
for specific mineralogical compositions, gel formation, time depen-
dency, among others, whose comprehension and prediction may lead to
benefits for the process (Bernhardt et al., 1999; Farrokhpay, 2012;
Vallar et al., 1999; Nguyen and Boger, 1998).
Rheology in pulps is dominated by physicochemical surface inter-
actions between particles and the fluid (Madigan et al., 2009; Zhou
et al., 1999, 2001). Whilst the most abundant components might be
expected to have the greatest impact (Johnson et al., 2000), the
sensitivity of rheological behavior to surface area can result in specific
minerals exerting an influence which is disproportionate with their
mass abundance. Clay minerals such as kaolinite and species belonging
to the smectite group (e.g. montmorillonite) are often a source of
ores (e.g. porphyry copper deposits (Maksaev et al., 2010; Alpers and
Brimhall, 1988) and their intrinsically small particle size/high surface
area (MacCarthy and Nosrati, 2013).
Among physicochemical surface interactions, according to DLVO
theory (Ninham, 1999), there are two main forces relevant to rheology:
electrostatic repulsion/attraction and Van der Waals attraction. The
first one exists due to the electrostatic double layer (EDL), which
comprises ions adsorbed to a particle surface, creating an energy barrier
that prevents other particles of like charge from colliding. This force is
very responsive to the concentration and speciation of solutes as well as
to the pH of the suspension. On the other hand, Van der Waals
attraction is caused by the distortion of the electron cloud as two
atoms approach one another, creating a dipole. This force predominates
whenever the electrostatic repulsion is weak (Boger, 2000; de Kretser
et al., 1998).
DLVO theory has been widely applied to the study of colloidal
stability of mineral suspensions. Some authors (Cruz and Peng, 2016)
claim that for low slurries with a low proportion of colloidal particles
relative to the total solids present, the rheological behavior is mostly
explained by the overall solid concentration. In the case of suspensions

⁎
Corresponding author.
E-mail address: javier.merrill@csiro.au (J. Merrill).

http://dx.doi.org/10.1016/j.mineng.2017.03.009
Received 30 June 2016; Received in revised form 22 November 2016; Accepted 25 March 2017
Available online 26 April 2017
0892-6875/ © 2017 Elsevier Ltd. All rights reserved.
J. Merrill et al. Minerals Engineering 109 (2017) 126–134

with high clay/colloidal contents the DLVO equation is highly relevant, Table 1
showing non-Newtonian flow that can have viscous, viscoelastic or Mixture design for each ternary plot, A, B and C being any three minerals from the five
selected.
elastic response. Moreover, EDL repulsion and Van der Waals attraction
effect upon the rheology of clay bearing suspensions has been acknowl- Mineral A (vol.%) Mineral B (vol.%) Mineral C (vol.%)
edged by multiple studies, through the relation observed between the
yield stress and the zeta potential due to pH (Goh et al., 2011) and other 100 – –
67.7 – 33.3
ions concentrations (Addai-Mensah, 2007).
33.3 – 67.7
Previous work (Ndlovu et al., 2014) indicates that suspensions – – 100
rheology is highly dependent on the specific clay species present in the 60 20 20
solid phase. As well, correlation has been found between the clay micro- 20 20 60
structures and rheological behavior, attributed in some part to the 67.7 33.3 –
33.3 33.3 33.3
natural shape and charge asymmetry of clay particles (Goh et al., 2011).
– 33.3 67.7
For example, bentonite slurries which are predominantly smectite in 33.3 67.7 –
composition, have been reported (Zhang and Peng, 2014; Cruz and 20 60 20
Peng, 2016; Zhang et al., 2015; Luckham and Rossi, 1999) to have – 67.7 33.3
– 100 –
greater influence on rheological properties when compared to kaolinite-
dominated slurries. This has been attributed to the face-edge and edge-
edge particles micro-structures, which contrast with kaolinite which
tend to form more compact face-face aggregates. Although the precise
nature of the physicochemical interactions of clay species is beyond the
scope of this investigation, the recognition of the complex interactions
involved plays a role in understanding rheological variability in ores
and synthetic mineral mixtures.
The present work studies the feasibility of estimating the rheological
behavior of slurries from mineral quantification models based on
hyperspectral data, which can be obtained at lower cost and time in
comparison to other characterization techniques such as XRD, SEM. The
ability to anticipate rheological behavior opens the possibility towards
developing efficient control systems for slurry holding and transport.

2. Experimental set-up and procedures

2.1. Methodology
Quartz (SiO2) and white mica were chosen as gangue minerals
because they are the main components in the feed of many Chilean
metallurgical plants. Bentonite and kaolin were used as sources of
montmorillonite and kaolinite, respectively.
Rheology flow curve measurements were conducted at a bulk solids
concentration by volume of 30% (54% in weight approx.). Due to the
high viscosity and yield stress of one of the bentonites, suspensions
were restricted to 15 vol.% (Blakey and James, 2003).
Hyperspectral characterization was carried out using the Hylogger3
(Huntington et al., 2010). The samples were briquettes made from dry
duplicates of the samples used for the suspensions. After thorough
mixing, the various mineral mixtures received no other treatment aside
from pressure before scanning. Several spectral measurements were
taken from each sample; some were used for creating models for
detecting the presence and the volumetric fraction of each of the five
minerals used, others were used for the evaluation of the models.
In order to build all ternary plots, a mixture design (Ayadi et al.,
2013) with 13 experiments was selected, as shown in Table 1 and Fig. 1.
Considering that the measurements for ternary plots containing the
high viscosity bentonite were carried out with 15 vol.% and the rest at
30 vol.%, and the fact that some vertices and edges were common to
some triple mixture ternary diagrams, a total of 81 rheology measure-
ments were made.
The density of the samples was determined using pycnometry.
Humidity tests were performed to determine moisture contents and the
exact amounts of mineral and water to be added when preparing the
slurries and the briquettes.
The same mixture design (Table 1 and Fig. 1) was used for the
hyperspectral characterization. Finally, using spectral features, mathe-
matical models have been developed for detecting the presence of any
of the five minerals, and once detected, another model was applied
towards estimating the abundance (volumetric fraction) of that specific
Fig. 1. Mixture design over ternary plot.
mineral in a sample, leading to a point on a ternary plot for viscosity
and another for yield stress obtained from a fitted Bingham Plastic
Model (Kwak et al., 2005).
2.2. Materials
Five minerals were used for this study: bentonite from a deposit in
northern Chile (Bentonite A), COMACSA® bentonite (Bentonite C)
COMACSA, 2016, ground COMACSA® white mica, COMACSA® kaolin
and finely ground quartz.
Quantitative X-ray diffraction analysis was conducted to determine
the mineralogical composition of each material used (Table 3). All XRD
data was collected under the same experimental conditions, in the
angular range 2° < 2θ < 80° using Bruker® D8 equipment. These
data compared favorably with hyperspectral analysis (see below) of the
samples.
To study the particle size distribution, a laser granulometry analysis
was performed for every material using Malvern® MasterSizer2000
equipment. These measurements were conducted in a wet environment
(distilled water) to replicate their true in-pulp size.
2.3. Preparation of slurries
Slurries were prepared by mixing specific amounts of minerals with
distilled and demineralized water. The pH of the synthetic slurries was
controlled by adding concentrated sulfuric acid or sodium hydroxide
solutions. The corresponding minerals were weighed according to the
requirements of each experiment. Subsequently, water was poured into
a measuring cylinder. The weighed solids were mixed into the
corresponding flask with the water and manually shaken until all solids

127
J. Merrill et al. Minerals Engineering 109 (2017) 126–134

were wet. Flasks with the slurries were placed into a mechanical stirring
device for 3 h at 200 RPM to ensure their homogeneity and disaggrega-
tion, after which each flask was withdrawn for pH adjustment and
subsequent rheological measurement. No pre-shearing has been made
protocol was not conducted due to buffer effects altering the pH of the
suspension. Instead, pH equivalence has been prioritized.
Neither the temperature effect, studied by other authors (Yang
et al., 2001), nor a temperature control mechanism was considered
herein, since the room was always kept between 21 and 23 °C, meaning
no significant perturbation for the experiments.

2.4. Moisture content

Fig. 2. Hylogger-3 measuring example scheme.
To estimate the moisture content in the different minerals, samples
were dried in a laboratory oven at 110 °C for 24 h. The difference in 2.7. Hyperspectral characterization
weight of the minerals after extraction from the oven corresponds to the
evaporated water. Hyperspectral characterization comprises the interpretation of the
spectral signature of light reflected by a sample, allowing the recogni-
tion of distinct chemical groups and their corresponding minerals,
2.5. Density estimation
mineral groups (and other substances where relevant). Samples are
exposed to a light source and a sensor measures the reflectance at
Each mineral was measured using a 100 ml pycnometer. The
different wavelengths (channels). This technique is attractive for the
corresponding results are shown in Table 2.
creation of a control system, due to its non-destructive nature, real time
results and low cost.
2.6. Rheometric measurements The spectral measurements were made with the Hylogger3 devel-
oped by CSIRO (2016), which operates within the VNIR (Visible Near
The rheometer used was an Anton Paar Rheolab QC using a Infrared: 350–1000 nm)-SWIR (Shortwave Infrared: 1000–2500 nm)
concentric cylinder geometry with 21 and 19.503 mm outer and inner and TIR (Thermal Infrared: 6000–14500 nm) spectral ranges. Hylog-
radio respectively. This instrument yields accurate measurements for ger3 takes one spectral measurement every 4 mm on a straight line over
high shear stress conditions, but rather noisy for low viscosity and low the samples, having six to seven pixels (measurements) per briquette.
yield stress materials. Moreover, concentric cylinder have problems Valid pixels were chosen according to certain albedo parameters,
determining the yield stress and shear stress at very low shear rate due signal-to-noise ratio and distance from edges to ensure reliable spectral
to wall slip (Kwak et al., 2005; He et al., 2004). Even though measurements were utilized in the further modelling. As shown in
sandblasted cylinder was used for minimizing this effect, the data in Fig. 2, three pixels per briquette were randomly chosen for the creation
the low ranges (0–100 s−1) was not taken into account during of the models for detection and quantification for each of the five
rheological model adjustment. minerals used. Valid pixels that were not used for the models develop-
It is well known that pH, water quality and solution speciation ment were applied in the error quantification and evaluation.
affects the rheological behavior of suspensions (Nosrati et al., 2012). In Hyperspectral analysis is a qualitative characterization tool, which
the present case, once the flasks with the contained slurries were means that the interpretation of the spectral data leads to the recogni-
removed from the agitation device, pH was adjusted to 7.0 in order to tion of minerals through their signature in the reflected signal.
modify the ionic strength as little as possible, since 7 is an intermediate Therefore, in the first instance it is not possible to render a quantitative
value between the natural pH of suspensions made of the 5 minerals account of abundances. Results shown in Table 3 are raw suggestions
utilized. The flask content was immediately poured into the rheometer made by TSA (The Spectral Assistant) algorithm (Berman and Bischof,
cup for measurement. Tests were carried out with a linear shear rate 2011), included in The Spectral Geologist, TSG software (CSIRO, 2014)
ramp between 0 and 300 s−1 in 1 min. In all measurements, a distinct used for the spectral data treatment.
linear region in the rheogram has been found, suitable for adjusting a
Bingham Plastic flow model (Boger, 1989): 3. Results and discussion
τ = τy + ηγ ,̇
3.1. Characterization
with τ , τy , η and γ ̇ the shear stress, yield stress, Bingham viscosity and
shear rate, respectively. The Bingham model was chosen because it A broad characterization of the materials used is fundamental
showed better adjustment than Herschel-Bulkley for high shear rates, towards the comprehension and discussion of the rheological phenom-
where a linear region was identified, while the Herschel-Bulkley model ena observed during the study. Tables 2, 3 and Figs. 3, 4 summarize the
adjusted better to low shear rate region, where measurements were characterization made for the five minerals.
assessed unreliable. As shown in Table 2, bentonites are highly hygroscopic. In the case
of these minerals, moisture was taken into account when calculating the
Table 2 amount of water to be added in order to reach the desired solids
Moisture, density and p80 from granulometry results. concentration, also precise densities are used for volume to mass
conversions.
Mineral Moisture Density (g/ml) d80 (µm)
Particle size is an important factor in rheology (Shi and Napier-
Bentonite C 10.20% 2.593 24.4 Munn, 1996). Table 2 and Fig. 3, showing granulometric results, reveal
White Mica 1.07% 2.728 38.9 that in general a considerable amount of these minerals is under 38 µm
Kaolin 0.88% 2.539 8.3
(No. 400 ASTM) in size.
Quartz 1.86% 2.630 62.4
Bentonite A 8.19% 2.739 40.7 Table 3 shows the results obtained from quantitative X-ray diffrac-
tion for each material. The X-ray diffraction shows expected results for

128
J. Merrill et al. Minerals Engineering 109 (2017) 126–134

Table 3
XRD and hyperspectral characterization results.

XRD HyLogger/TSA

Mineral Major Minor Trace SWIR TIR

Bentonite C Amorphous Anorthite Cristobalite Muscovite Montmorillonite Montmorillonite

Quartz Pyrophyllite
Microcline
White Mica Muscovite – – Muscovite Muscovite
Magnesite
Kaolin Kaolinite Quartz Muscovite Dickite Kaolinite
Dickite Alunite Pyrophyllite Kaolinite Pyrophyllite
Calcite
Anhydrite
Anatase
Amorphous
Quartz Quartz – – – Quartz
Alunite-Na Diopside
Bentonite A Andesine Kaolinite – Montmorillonite Montmorillonite
Muscovite Quartz Kaolinite Albite
Amorphous Orthoclase
Anhydrite

differed in the quantity and speciation of their accessory mineral

Fig. 3. Full granulometry results: retained fraction vs. size.
content.
Hylogger3 analysis has a good correlation with XRD results, but
these are just qualitative results that correspond to the un-mixing
algorithm (Berman and Bischof, 2011), suggesting a mineral mixture
from spectral measurements based on the TSG software mineral library.
Fig. 4 presents the spectra measured for pure bentonite A, where
main spectral features have been marked at 1414, 1912 and 2206 nm of
wavelength based on the montmorillonite reference from the TSG
mineral library. A very good match with such references is exhibited,
thus giving a strong indication of its presence.
Material properties shown in Table 2 correlates well with the
characterization conducted. As expected, the two bentonites stood out
in moisture values because of their smectite group mineral content,
although bentonite C exceled due to its smaller particle size, therefore
higher net surface area.
As well, densities can be explained by the minerals detected, for
example; montmorillonite having an average density of 2.35 g/cm3, in
bentonite A is mixed with andesine (2.67 g/cm3) and muscovite
(2.82 g/cm3), and in bentonite C with anorthite (2.73 g/cm3).

3.2. Rheometric results

The quality of the rheometric results was highly variable. As shown

in Fig. 5, bentonite A has very high stress conditions and a stable
rheogram in comparison with quartz in Fig. 6. Considering the same
measurement conditions, the quartz rheogram is very irregular and
noisy. In these cases, the Bingham model adjustment was made with the
best segment available, for example, in the case of quartz, the shear rate

Fig. 4. Spectral sample of bentonite A. Dashed lines mark montmorillonite main spectral
features.

the white mica, quartz and kaolin, but seems to have problems
detecting montmorillonite, maybe confusing it with an amorphous
phase, but hyperspectral analysis confirmed the presence of montmor-
illonite in both bentonites.
The XRD characterization of the materials revealed that the white
mica was primarily muscovite (KAl2(Si3Al)O10(OH,F)2), whilst both
dickite and kaolinite (both Al2(Si2O5)(OH)4), along with minor quartz
and pyrophyllite were identified in the kaolin. Both bentonite samples
contained montmorillonite (Na,Ca)0.3(Al,Mg)2Si4O10(OH)2·nH2O, but
Fig. 5. Rheogram for bentonite A at 30 vol.%, pH 7.00.

129
J. Merrill et al. Minerals Engineering 109 (2017) 126–134

Fig. 6. Rheogram for quartz at 30 vol.%, pH 7.08.

would be between 100 and 200 s−1, giving a viscosity of 6.2 mPa·s and
Fig. 7. Viscosity ternary plot in vol. fraction for bentonite A, bentonite C and white mica
a yield stress of 0.16 Pa. Note that in this case the yield stress is close to at 15% Cv and pH 7.
0, resembling a Newtonian fluid.
After adjusting a Bingham Plastic flow model for each one of the 81
rheometry measurements, the results were plotted in ternary contour
diagrams, where sometimes counter-intuitive results, such as sharp
increases in viscosity and yield stress and other synergic behaviors
between minerals, are observed. From all 20 ternary plots created, two
were chosen for this work due to their outstanding non-linear mixing
effects. This corresponding phenomena is discussed in the following
sections of this work.
Table 4 presents the parameters for the Bingham model for the five
minerals used. A steep increment in Bingham viscosity and yield stress
from 15% to 30% Cv as slurries get closer to the maximum fraction of
packing is found (Alejo and Barrientos, 2009). It is noted that
rheological tests could not be performed to pure bentonite C at solids
concentrations above 10% Cv, because of the high viscosity and yield
stress. In particular, the slurry could not be poured from the flask to the
cup.
Even in small amounts, bentonite C caused a significant increase in Fig. 8. Viscosity ternary plot in vol. fraction for white mica, kaolin and quartz at 30% Cv
viscosity and yield stress, as previously found for bentonite concentra- and pH 7.
tions above 6% by weight (Bekkour et al., 2005). As indicated by Fig. 7.
Unexpectedly, bentonite A was the only mineral that dropped this sharp of suspensions with the same minerals alone. In this case, a slurry at
rise, even when there is more smectite minerals added to the pulp. This 30% Cv with pure white mica and pure kaolin, showed higher viscosity
suggests that even a minor amount of impurities in this mineral exerts a values that when combined, reaching a minimum when the ratio of
rheology modification mechanism for bentonite, being orthoclase, kaolin and white mica is 2/3 respectively.
anhydrite and andesine the candidates for this behavior or maybe a This may be explained by structure formation of white mica
synergic effect between two or more of them. Discarding kaolin and particles while bonding each other, which is cause of strong time
muscovite can take place, since these minerals were present in the other dependency for rheological properties and shear thinning behavior
materials used in this study, and no other steep drops of the rheological (Nosrati et al., 2011). Kaolin might affect these structures because of its
parameters were identified for bentonite C. fine particle size and the edge-face charge heterogeneity of the kaolinite
Another synergistic effect can be seen between white mica and particles. Quartz does not affect this behavior, keeping the minimum
kaolin (Fig. 8). In most cases, the viscosity of a suspension with viscosity at a kaolin-white mica ratio of 2/3, reaffirming the hypothesis
combined minerals has an intermediate value between the viscosities of structures being affected by electrostatic unbalances, having silica a
well-balanced crystal structure. Another hypothesis suggests that this
non-intuitive rheological behavior relies on competitive ion adsorption
Table 4
(Goldberg, 2002) due to the presence of more than one species,
Bingham model parameters adjusted for ternary vertices (pure minerals).
generating non-DLVO interactions. Synergistic effects were not ob-
Cv 30% Cv 15% served in the corresponding Bingham yield stress ternary plot, but since
Bingham flow model is inaccurate for the low shear rate region for most
Viscosity Yield stress Viscosity Yield stress suspensions in this study, it is not conclusive.
(mPa·s) (Pa) (mPa·s) (Pa)
The knowledge of the aforementioned effects could lead to bene-
Bentonite A 296.1 146.7 5.9 0.87 ficial decision making in a processing plant, allowing to control high
Bentonite C – – 440.7a 293.6a viscosities and yield stress in case they are present, possibly, with the
White Mica 52.6 14.4 6.6 0 minerals available in the mineral deposit itself.
Kaolin 57.5 35.1 5.7 1.21
Putting these features in the context of a slurry transport system,
Quartz 6.2 0.16 5.0 0
present results expose the potential to reduce water consumption via
a
For plotting purposes the missing value was replaced by 1.1 times the maximum raising the solids concentration, maintaining rheological properties
value measured, that corresponded to 400.63 mPa·s for viscosity and 266.94 Pa for yield within desired ranges, increasing the treatment capacity of the unit
stress, reached by the 66.67% bentonite C and 33.33% white mica mixture slurry, giving processes.
440.693 and 293.63 Pa for the missing values.

130
J. Merrill et al. Minerals Engineering 109 (2017) 126–134

For the comprehension of synergic rheological effects between

minerals in suspensions, further research must be carried out, aiming
to understanding the exact mechanisms of interaction, focusing on
surface physicochemical properties such as zeta potential (Greenwood,
2003) and crystal structure charge balance. Also, these results consider
the case of pH = 7, the expansion of this work to other pH values is
proposed, since variations on the pH have a strong impact on results.

3.3. Abundance models based on hyperspectral analysis results

Spectral features like troughs, peaks, “shoulders”, and their para-

meters like depth, height, asymmetry, position, etc., may be used for
recognizing minerals, but it is also possible to find a correlation
between them with mineral abundance. Several features were studied
with the aim of developing models towards the recognition of the
presence of a mineral and to estimate its abundance, sometimes, but not
always, being the same feature for both purposes.
Regarding mineral recognition, the intent is to find unique spectral
features of that specific mineral. For quantification purposes, a para-
meter of a spectral feature requires having a good correlation with
abundance. Spectral features are analyzed by TSG software using
“scalars”, which are an arrangement of numbers obtained for every
sample, through mathematical spectral data treatment, such as a
polynomial fit, in order to obtain a single numerical value that
represents a parameter from the aimed spectral feature.

3.3.1. Detection and quantification modelling

White mica shares spectral features with kaolin and the two
bentonites, although it is recognizable by the height and exact position
of a peak in the TIR range, around 9613 nm wavelength. On the other
hand, a very poor correlation with abundance is obtained. With the
parameters shown in Table 5 there were 4 false negatives, meaning that
there was white mica in the samples and it was not detected, in a total
of 447 measurements, showing robustness in the recognition model.
Modifying the values of the recognition parameters, the number of false
positives and false negatives is altered, for example allowing the
selection of the best parameters, according to the beneficial behavior
Fig. 9. Spectral measurement for: (a) quartz-bentonite C-bentonite A, (b) pure kaolin and
expected from a process (either overestimating or underestimating). (c) pure white mica samples. Dashed line encloses 2300–2500 nm area.
A model for the abundance of white mica was developed, using a set
of troughs between 2300 nm and 2500 nm, using the energy absorption MCscalar2 and the white mica abundance giving the model shown in
between those wavelengths, but since kaolin also has some spectral
Table 6.
features in that range, a kaolin abundance model had to be included in Here x is the energy absorption in the 2300 and 2500 nm
order to differentiate the effect of the two minerals upon the measured
wavelength range.
spectra (see Fig. 9). Finally, for mixtures, after every model detected and estimated the
Fig. 10 shows the correlation between the energy absorption in the abundance of its corresponding mineral, a normalization of the values
2300 to 2500 nm wavelength region, represented as spectral absorption
was carried out for them to sum 1.
area, and the volumetric fraction of the white mica, the color of the dots Different models were used according to the nature of the correla-
shows a deviation for high kaolin abundance. This deviation may be
tions, finally giving an acceptable Mean Absolute Deviation (MAD) for
corrected by using some spectral feature correlated to kaolin abun- the five models, as shown in Table 7. The worst case was the white mica
dance, which was the depth of a trough at a wavelength of 1380 nm
model, where, in average, the model indicated a volumetric fraction
exhibited by the samples containing kaolin. 3.25% far from the real abundance, being a good estimation for
After making the aforementioned mathematical adjustment, the
rheological purposes.
new scalar, named MCscalar2, was plotted against white mica real
abundance in Fig. 11, showing better correlation. Leaving out the
samples with no white mica, a linear adjustment was made between de

Table 5
Mineral recognition parameters and errors in 447 samples.

Mineral Spectral feature Height/deepness limit Position limit False negatives False positives

White Mica Peak at 9613 nm ≥0.0212 ≤9627.38 nm 4 0

Bentonite C Trough at 1910 nm ≥0.0961 ≥1905.11 nm 1 0
Bentonite A Trough at 1910 nm ≥0.143 ≤1909.62 nm 0 0
Kaolin Trough at 2200 nm ≥0.0704 – 0 0
Quartz Peak at 8540 nm ≥0.0237 – 0 0

131
J. Merrill et al. Minerals Engineering 109 (2017) 126–134

Linear viscosity model

log(η) = 1.658 + 0.054X1−0.859X3

Non-linear yield stress model

τ = 33.451−0.319X1−2.193X3−0.65X1 X2−0.771X1 X3 + 2.277X2 X3−3.058X1 X2

X3

Non-linear viscosity model

log(η) = 1.764−0.028X1−0.989X3−0.619X1 X2 + 0.313X1 X3−0.279X2 X3

+ 0.317X1 X2 X3

Fig. 10. White mica abundance (volumetric fraction) vs. Energy absorption in the 2300
3.4.1. Model evaluation
and 2500 nm wavelength range. The first evaluation process was to compare values given by linear
and non-linear models, using the real abundances, and then plotting the
results in ternary diagrams for visual comparison. Table 8 shows the
real abundances for the three minerals, along with measured and
modelled viscosity and yield stress.
In Fig. 12, the linear model shows a loss of curvature, occluding the
synergic effect between kaolin and white mica, while the non-linear
model shows a good fit.
For yield stress, Fig. 13 shows a large overestimation for high
values, again showing a better fit in the non-linear model.
Finally, hyperspectral data was used as an input for linear and non-
linear models towards calculating the Mean Absolute Deviation (MAD)
of the rheological properties estimation. Results are shown in Table 9.
MAD values are lower for non-linear models. These values are small
in comparison with the high viscosity or yield stress values reached by
slurries, indicating an appropriate estimation process starting from
hyper-spectral characterization and finally obtaining estimated rheolo-
gical properties.
Fig. 11. White mica abundance vs. MCscalar2. This scheme of rheology control model might be adopted in an
online hyperspectral sampling of the crusher output, towards improved
Table 6 decision making in real time to optimize the use of water or additives
Linear model fitted for white mica volumetric fraction.
for the milling or other downstream processes. Alternatively, applica-
Linear model fitted m c R2 tion of the approach in geometallurgical models might enable improved
feasibility studying, when considering predictions of rheological prop-
y = mx + c 0.2675 −0.0485 0.952 erties for the slurries that will be treated, while at the same time,
making profitable the extraction of metals from deposits with high
contents of clay minerals, through the incorporation of rheology control
Table 7
systems.
Mean absolute deviation.

Mineral MAD (vol.%)

4. Conclusions
Bentonite A 2.03
Bentonite C 2.04 A set of rheology measurements was carried out for synthetic
White Mica 3.25
Kaolin 2.12
slurries with controlled composition, obtaining an array of ternary
Quartz 1.90 plots for Bingham yield stress and viscosity according to mineral
mixtures, where two significant synergistic effects were found. The
first one was a minimum viscosity found for a kaolin/white mica ratio
of 2:3, where each mineral on its own had over 50 mPa·s viscosity, but
3.4. Ternary plots modelling combined in such ratio, it was lowered to 38 mPa·s for a suspension
with 30% Cv solids concentration. The other observed effect was a
For modelling rheological properties (Bingham yield stress and viscosity decreasing of a 15% Cv solids concentration bentonite
viscosity) from mineralogical composition, linear and non-linear mod- suspension (over 400 mPa·s) by adding bentonite from another source.
els were fitted for white mica – kaolin – quartz ternary plots (viscosity Finally, hyperspectral analysis in the VNIR-SWIR and TIR region
shown in Fig. 8). was conducted for a duplicated solid phase of the slurries, enabling the
The following equations are the fitted models, where X1, X2 and X3 development of abundance quantification models for the minerals
correspond to volumetric fractions of white mica, kaolin and quartz studied, which performed well, giving accurate estimates of the samples
respectively. composition. Yield stress and viscosity calculated from mineral compo-
Linear yield stress model sition showed low discrepancies with measurements. Slurry rheological
parameters estimation using hyperspectral data has proven to be
log(τ ) = 1.75−0.529X1−2.177X3
feasible.

132
J. Merrill et al. Minerals Engineering 109 (2017) 126–134

Table 8
Data and results for fitted viscosity and yield stress models.

X1 X2 X3 Measured Linear Non-linear

White Mica (vol.%) Kaolin (vol.%) Quartz (vol.%) τ (Pa) η (mPa·s) τ (Pa) η (mPa·s) τ (Pa) η (mPa·s)

100 14.39 52.55 16.65 51.54 15.01 54.38

100 35.13 57.53 56.26 45.53 33.45 58.05
100 0.16 6.20 0.37 6.30 1.16 5.96
33 67 29.57 43.90 37.49 47.45 32.58 41.38
33 67 1.30 12.52 1.33 12.69 0.70 14.62
67 33 30.39 37.27 24.98 49.45 26.43 40.49
33 67 4.17 11.82 1.99 12.18 3.68 11.04
67 33 4.35 35.83 4.70 25.58 5.31 30.54
67 33 13.60 22.03 10.58 23.55 14.44 23.57
60 20 20 10.10 33.59 9.94 33.02 11.56 32.27
20 60 20 16.28 31.27 16.18 31.42 17.80 29.71
33 33 33 8.43 22.60 7.05 24.54 7.90 23.53
20 20 60 3.32 13.78 2.18 14.25 1.16 14.19

Table 9
Mean Absolute Deviation for rheological properties using abundances from hyperspectral
quantification models in linear and non-linear models for viscosity and yield stress.

Yield stress Viscosity

MAD (Pa) MAD (mPa·s)

Linear 4.04 3.34

Non-linear 1.78 1.69

Acknowledgements

The authors of this research work acknowledge the AMTC, CSIRO

Chile and the Universidad de Chile institutions. This research was
performed within the framework of the project INNOVA CORFO
10CEII-9007. CFI gratefully acknowledges support from the National
Commission for Science and Technology (Conicyt) through Fondecyt
Project 1160971.

References

Fig. 12. White mica - kaolin - quartz viscosity plot for measured, linear and non-linear
models fitted.
Fig. 13. White mica - kaolin - quartz yield stress plot for measured, linear and non-linear
models fitted.
Addai-Mensah, J., 2007. Enhanced flocculation and dewatering of clay mineral
dispersions. Powder Technol. 179, 73–78. http://dx.doi.org/10.1016/j.powtec.2006.
11.008.
Alejo, B., Barrientos, A., 2009. Model for yield stress of quartz pulps and copper tailings.
Int. J. Miner. Process. 93, 213–219. http://dx.doi.org/10.1016/j.minpro.2009.08.
002.
Alpers, C.N., Brimhall, G.H., 1988. Middle Miocene climatic change in the Atacama
Desert, northern Chile: evidence from supergene mineralization at La Escondida. Bull.
Geol. Soc. Am. 100, 1640–1656. http://dx.doi.org/10.1130/0016-7606(1988)
100<1640:MMCCIT>2.3.CO;2.
Ayadi, A.J., Soro, J., Kamoun, A., Baklouti, S., Sfax, R.D.S., Thomas, A.A., 2013. Study of
clay’s mineralogy effect on rheological behavior of ceramic suspensions using an
experimental design. Int. J. Res. Rev. Appl. Sci. 14, 374–384.
Bekkour, K., Leyama, M., Benchabane, A., Scrivener, O., 2005. Time-dependent
rheological behavior of bentonite suspensions: an experimental study. J. Rheol. (N. Y.
N. Y) 49, 1329–1345. http://dx.doi.org/10.1122/1.2079267.
Berman, M., Bischof, L., 2011. An unmixing algorithm based on a large library of
shortwave infrared spectra. Tech. Rep.. https://publications.csiro.au/rpr/
download?pid=csiro:EP117468&dsid=DS4 (accessed March 21, 2016).
Bernhardt, C., Reinsch, E., Husemann, K., 1999. The influence of suspension properties on
ultra-fine grinding in stirred ball mills. Powder Technol. 105, 357–361. http://dx.doi.
org/10.1016/S0032-5910(99)00159-X.
Blakey, B.C., James, D.F., 2003. Characterizing the rheology of laterite slurries. Int. J.
Miner. Process. 70, 23–39. http://dx.doi.org/10.1016/S0301-7516(02)00147-3.
Boger, D.V., 1989. An introduction to rheology. J. Nonnewton. Fluid Mech. 32, 331–333.
http://dx.doi.org/10.1016/0377-0257(89)85015-3.
Boger, D.V., 2000. Rheology and the minerals industry. Miner. Process. Extr. Metall. Rev.
20, 1–25. http://dx.doi.org/10.1080/08827509908962460.
COMACSA, COMACSA Cía. Minera Agregados Calcáreos S.A., 2016. < http://www.
comacsa.com.pe/ > (accessed April 1, 2016).
Cruz, N., Peng, Y., 2016. Rheology measurements for flotation slurries with high clay
contents - a critical review. Miner. Eng. 98, 137–150. http://dx.doi.org/10.1016/j.
mineng.2016.08.011.

133
J. Merrill et al.
Cruz, N., Peng, Y., Farrokhpay, S., Bradshaw, D., 2013. Interactions of clay minerals in
copper-gold flotation: Part 1 - Rheological properties of clay mineral suspensions in
the presence of flotation reagents. Miner. Eng. 50–51, 30–37. http://dx.doi.org/10.
1016/j.mineng.2013.06.003.
CSIRO, The Spectral Geologist, 2014. <http://www.thespectralgeologist.com/>
(accessed June 9, 2015).
CSIRO, Commonwealth Scientific and Industrial Research Organisation, Australian
Government, 2016. < http://www.csiro.au/ > (accessed April 1, 2016).
de Kretser, R.G., Scales, P.J., Boger, D.V., 1998. Surface chemistry-rheology inter-
relationships in clay suspensions. Colloids Surf. A Physicochem. Eng. Asp. 137,
307–318. http://dx.doi.org/10.1016/S0927-7757(97)00372-5.
Farrokhpay, S., 2012. The importance of rheology in mineral flotation: a review. Miner.
Eng. 36–38, 272–278. http://dx.doi.org/10.1016/j.mineng.2012.05.009.
Goh, R., Leong, Y.K., Lehane, B., 2011. Bentonite slurries-zeta potential, yield stress,
adsorbed additive and time-dependent behaviour. Rheol. Acta. 50, 29–38. http://dx.
doi.org/10.1007/s00397-010-0498-x.
Goldberg, S., 2002. Competitive adsorption of arsenate and arsenite on oxides and clay
minerals. Soil Sci. Soc. Am. J. 413–421. https://dl.sciencesocieties.org/publications/
sssaj/abstracts/66/2/413 (accessed January 12, 2016).
Greenwood, R., 2003. Review of the measurement of zeta potentials in concentrated
aqueous suspensions using electroacoustics. Adv. Colloid Interface Sci. 106, 55–81.
http://dx.doi.org/10.1016/S0001-8686(03)00105-2.
He, M., Wang, Y., Forssberg, E., 2004. Slurry rheology in wet ultrafine grinding of
industrial minerals: a review. Powder Technol. 147, 94–112. http://dx.doi.org/10.
1016/j.powtec.2004.09.032.
Huntington, J., Whitbourn, L., Mason, P., Berman, M., Schodlok, M.C., 2010. HyLogging -
Voluminous industrial-scale reflectance spectroscopy of the earth’s Subsurface. In:
Proc. ASD IEEE GRS; Art, Sci. Appl. Reflectance Spectrosc. Symp., Boulder, CO Vol. II.
pp. 1–14. 14pp. www.asdi.com. http://www.asdi.com/getmedia/364fb0a4-ad5f-
4e3d-b6e7-76ccb7d4038f/HyLogging.pdf.aspx.
Johnson, S.B., Franks, G.V., Scales, P.J., Boger, D.V., Healy, T.W., 2000. Surface
chemistry–rheology relationships in concentrated mineral suspensions. Int. J. Miner.
Process. 58, 267–304. http://dx.doi.org/10.1016/S0301-7516(99)00041-1.
Kelm, U., Helle, S., 2005. Acid leaching of malachite in synthetic mixtures of clay and
zeolite-rich gangue. An experimental approach to improve the understanding of
problems in heap leaching operations. Appl. Clay Sci. 29, 187–198. http://dx.doi.
org/10.1016/j.clay.2004.12.004.
Klein, B., Hallbom, D., 2002. Modifying the rheology of nickel laterite suspensions. Miner.
Eng. 15, 745–749. http://dx.doi.org/10.1016/S0892-6875(02)00170-X.
Kwak, M., James, D.F., Klein, K.a., 2005. Flow behaviour of tailings paste for surface
disposal. Int. J. Miner. Process. 77, 139–153. http://dx.doi.org/10.1016/j.minpro.
2005.06.001.
Luckham, P.F., Rossi, S., 1999. Colloidal and rheological properties of bentonite
suspensions. Adv. Colloid Interface Sci. 82, 43–92. http://dx.doi.org/10.1016/
S0001-8686(99)00005-6.
MacCarthy, J., Nosrati, A., 2013. Influence of mineral composition on rheology and acid
leaching behaviour of mixed oxide and clay minerals. Chemeca 41, 448–453.
<http://search.informit.com.au/
documentSummary;dn=882161946797435;res=IELENG> (accessed March 21,
2016).
134
Minerals Engineering 109 (2017) 126–134
Madigan, C., Leong, Y.K., Ong, B.C., 2009. Surface and rheological properties of as-
received colloidal goethite (α-FeOOH) suspensions: pH and polyethylenimine effects.
Int. J. Miner. Process. 93, 41–47. http://dx.doi.org/10.1016/j.minpro.2009.05.005.
Maksaev, V., Almonacid, T.a., Munizaga, F., Valencia, V., McWilliams, M., Barra, F.,
2010. Geochronological and thermochronological constraints on porphyry copper
mineralization in the Domeyko alteration zone. Andean Geol. 37, 144–176. http://
dx.doi.org/10.4067/S0718-71062010000100007.
Ndlovu, B., Forbes, E., Farrokhpay, S., Becker, M., Bradshaw, D., Deglon, D., 2014. A
preliminary rheological classification of phyllosilicate group minerals. Miner. Eng.
55, 190–200. http://dx.doi.org/10.1016/j.mineng.2013.06.004.
Nguyen, Q.D., Boger, D.V., 1998. Application of rheology to solving tailings disposal
problems. Int. J. Miner. Process. 54, 217–233. http://dx.doi.org/10.1016/S0301-
7516(98)00011-8.
Ninham, B.W., 1999. On progress in forces since the DLVO theory. Adv. Colloid Interface
Sci. 83, 1–17. http://dx.doi.org/10.1016/S0001-8686(99)00008-1.
Nosrati, A., Addai-Mensah, J., Skinner, W., 2011. Rheology of aging aqueous muscovite
clay dispersions. Chem. Eng. Sci. 66, 119–127. http://dx.doi.org/10.1016/j.ces.2010.
06.028.
Nosrati, A., Addai-Mensah, J., Skinner, W., 2012. Rheological behavior of muscovite clay
slurries: effect of water quality and solution speciation. Int. J. Miner. Process.
102–103, 89–98. http://dx.doi.org/10.1016/j.minpro.2011.09.009.
Shi, F.N., Napier-Munn, T.J., 1996. A model for slurry rheology. Int. J. Miner. Process. 47,
103–123. http://dx.doi.org/10.1016/0301-7516(95)00101-8.
Tan, H., Skinner, W., Addai-Mensah, J., 2012. Leaching behaviour of low and high Fe-
substituted chlorite clay minerals at low pH. Hydrometallurgy 125–126, 100–108.
http://dx.doi.org/10.1016/j.hydromet.2012.05.015.
Tan, H., Skinner, W., Addai-Mensah, J., 2013. Influence of fluorite on the isothermal
leaching and rheological behaviours of chlorite mineral pulps at low pH. Int. J.
Miner. Process. 123, 1–8. http://dx.doi.org/10.1016/j.minpro.2013.04.018.
Vallar, S., Houivet, D., El Fallah, J., Kervadec, D., Haussonne, J.-M., 1999. Oxide slurries
stability and powders dispersion: optimization with zeta potential and rheological
measurements. J. Eur. Ceram. Soc. 19, 1017–1021. http://dx.doi.org/10.1016/
S0955-2219(98)00365-3.
Wang, Y., Peng, Y., Nicholson, T., Lauten, R.A., 2015. The different effects of bentonite
and kaolin on copper flotation. Appl. Clay Sci. 114, 48–52. http://dx.doi.org/10.
1016/j.clay.2015.05.008.
Yang, H.-G., Li, C.-Z., Gu, H.-C., Fang, T.-N., 2001. Rheological behavior of titanium
dioxide suspensions. J. Colloid Interface Sci. 236, 96–103. http://dx.doi.org/10.
1006/jcis.2000.7373.
Zhang, M., Peng, Y., 2014. Effect of clay minerals on pulp rheology and the flotation of
copper and gold minerals. Miner. Eng. 70, 8–13. http://dx.doi.org/10.1016/j.
mineng.2014.08.014.
Zhang, M., Peng, Y., Xu, N., 2015. The effect of sea water on copper and gold flotation in
the presence of bentonite. Miner. Eng. 77, 93–98. http://dx.doi.org/10.1016/j.
mineng.2015.03.006.
Zhou, Z., Solomon, M.J., Scales, P.J., V Boger, D., 1999. The yield stress of concentrated
flocculated suspensions of size distributed particles. J. Rheol. (N. Y. N. Y). 43.
Zhou, Z., Scales, P.J., Boger, D.V., 2001. Chemical and physical control of the rheology of
concentrated metal oxide suspensions. Chem. Eng. Sci. 56, 2901–2920. http://dx.doi.
org/10.1016/S0009-2509(00)00473-5.