You are on page 1of 29

Quantitative analysis of iron ore

using SEM-based technologies 5


I.. Toneti
Indigo Research Laboratories, Johannesburg, South Africa

5.1Introduction
Improved characterization of lower-grade iron ores is becoming more critical owing
to the depletion of world high-grade iron ore reserves. The mineralogy and gangue
minerals of the lower-grade iron ores are found to affect the quality of pretreated ag-
glomerates of fine ore (sinters and pellets), which are making up a growing proportion
of blast-furnace feedstock (currently ~70 % in East Asia).
The understanding of lower-grade iron ore mineralogy and the subsequent char-
acterization of lower-grade iron ore are therefore becoming increasingly important
to the efficient processing of the ores and the optimization of downstream smelting
processes.
Using scanning electron microscope (SEM)-based technologies to characterize and
quantitatively analyze iron ores and iron ore mineralogy dates back to the invention
of the technology in the late 1950s to the early 1960s. The emphasis shifted from the
chemical analysis of penalty elements on microprobe systems to quantitative mineral-
ogy on automated mineralogical systems (auto-SEMs) only relatively recently (early
1990s). However, auto-SEMs consistently struggled to differentiate between the most
common iron oxides present within iron ore, those being magnetite (Fe3O4) and he-
matite (Fe2O3), since auto-SEMs traditionally used energy-dispersive spectroscopic
(EDS) technology for Fe quantification and furthermore ran at low X-ray counts.
Assay reconciliation, by comparing the amount of Fe measured from chemical analy-
ses against the calculated amount of Fe in a sample based on its mineralogy, was thus
essential for these systems, in gaining some kind of grasp on the amount of hematite/
magnetite in an ore body up until about 2008 and often required further X-ray diffrac-
tion (XRD) analysis for confirmation, which with the advent of Rietveld analysis has
become more quantitative in recent years.
Improved beam, system, and vacuum stabilities and in some instances stretched
brightness/contrast signals resulted in better, more consistent backscattered electron
(BSE) imaging, which allowed for better magnetite/hematite discrimination. With the
arrival of an appropriate level of hematite/magnetite distinction came the ability to use
the input of theoretical densities to predict the performance of various mechanisms of
density separation, mostly sink-float or washability analysis (B. Ntsoelengoe, 2011,
personal communication). Also, with an appreciation of the different chemistries dis-
tinguishing acicular SFCAs from columnar/tabular SFCAs, the analysis of iron ore sin-
ters in these systems was relooked at from about 2010 (Toneti and Dippenaar, 2011).

Iron Ore. http://dx.doi.org/10.1016/B978-1-78242-156-6.00005-8


2015 Elsevier Ltd. All rights reserved.
162 Iron Ore

With this came the realization that auto-SEMs, in one form or another, could be used
for mineralogical quantification through the project flow cycle, from exploration
(where bulk ore mineralogy was most important), to mining (recently geometallurgi-
cal modeling), to processing (concentrator/mining products), to utilization (sinter and
pellets; Table5.1). Auto-SEMs even found application in the characterization of iron
ore mine overburden, consisting mostly of pedocrete (Toneti, 2012). This proved
vital since it had a direct bearing on the calculations of the amount of overburden
occurring over the deposit, which in turn related directly to the cost of the overburden
removal (usually calculated on mass/density and not volume). Of specific note was
the occurrence of swelling clays (smectite) that had a direct bearing on the volume of
overburden encountered, though XRD is still a better technique for clay mineral spe-
ciation since many clay minerals have similar chemistry but different lattice structures.

5.2Principles of SEM-based technologies


5.2.1Introduction to the principle of scanning electron
microscopy
The scanning electron microscope (often abbreviated SEM) is the electron beam ana-
log of a reflected light microscope. Quantum theory dictates that all matter/energy ex-
hibits a dual nature, possessing both wave and particle (quantum) properties. It should
then come as no surprise that electrons can be used in a very similar manner to light
by creating an electron beam from an electron source (called an electron gun) in
a vacuum environment. The relationship of particles (in this instance, electrons) to
waves can be expressed through the de Broglie wavelength equation:

l = h /p

where is the wavelength, h is Planck's constant, and p is the momentum


(massvelocity) of the traveling body/quanta.
All this demonstrates that electrons can be refracted, reflected, absorbed, and trans-
mitted just like light.
In addition, primary electrons (from the electron gun) can generate characteristic
X-rays from electrons being displaced in the atomic orbitals of the sample being
analyzed (conservation of energy principle), which can be detected with X-ray de-
tectors and are characteristic of chemical elements present within the minerals be-
ing analyzed. The electrons that are reflected are called backscattered electrons
(Figure5.1). Generated secondary electrons (primarily a surface phenomenon) can
also be used to generate images (but not for any kind of mineralogical quantifica-
tion). Even though certain iron ore minerals may cathodoluminescence (give off
light) in the presence of an electron beam, cathodoluminescence is not currently
employed for iron ore mineral characterization. It may however provide a further
means to separate distinct grains from the same minerals, which happen to be touch-
ing, in the future.
Quantitative analysis of iron ore using SEM-based technologies163
Table5.1 The possible uses of auto-SEMs in a project lifecycle
Ore characterization/ Mining/concentrator
Project flow Exploration geometallurgical modeling products Sinters/pellets
Auto-SEM use Bulk mineralogy Density separation Bulk Mineralogy SFCAa Differentiation
Microlithotyping Sink-float analysis/washability Liberation Analysis Bulk Mineralogy
prediction (grade-recovery)
Grain size analysis Bulk mineralogy Grain Size Analysis
Textural analysis Bandwidth determination Fe Deportment
Fe deportment Clast size determination Locking Analysis

a
Silicoferrite of calcium and aluminum, a product of sintering low-grade fine iron ore to create a feed for blast furnaces. For more information, see Toneti and Dippenaar (2011).
164 Iron Ore

Primary
Backscattered electron beam
electrons
Characteristic
X-rays

Secondary Cathodoluminescence
electrons

Surface charge (a.k.a.


Absorbed
specimen current)
electrons

Transmitted
electrons

Figure5.1 Diagram of electron beamspecimen interaction.

To summarize, the SEM uses essentially two characteristics of materials to identify


minerals.

5.2.1.1The average atomic number of a mineral or substance


This translates to its BSE grayscale value (in the range 0255). Occasionally, it is
heard in industry that this relates to the density of a mineral although this is not tech-
nically true, since magnetite appears brighter in SEMs than hematite, even though it
is less dense. It does however have a slightly higher average atomic number (compare
the ratio of Fe to O in the two minerals and one will readily see this to be the caseFe
having a higher atomic number than O). The confusion arises out of very dense sub-
stances typically appearing white in analyses or calibration procedures.

5.2.1.2Characteristic X-ray spectra of minerals


It is the measurement of characteristic chemical X-ray spectra that makes any SEM
(with attached X-ray detectors) essentially a chemical tool that uses stoichiometry (the
measurement of characteristic chemical ratios in compounds and mineralsalbeit ap-
proximately) to characterize minerals and materials.

5.2.2Sample preparation principles


Conventional sample preparation techniques include the setting of samples in epoxy
resin within (typically) 30mm diameter mounts. The addition of graphite ensures par-
ticle separation. Depending on the sample holder and instrument type, 716 samples
can be accommodated at any one time. Larger rock chips and even cores or slabs have
been known to be measured in these systems; however, these are often related to on-
site special core/rock/slab holder engineering capabilities. In the case of slabs and drill
Quantitative analysis of iron ore using SEM-based technologies165

core, since most iron ore is sedimentary in origin, great care must be taken to ensure
that the area for cutting is in profile across bedding planes to maximize the discrimi-
nation of any textural features.

5.2.3Various SEM-based technologies


5.2.3.1Electron probe microanalysis
In the late 1940s, focused static electron beams or probes (with the aid of opti-
cal microscopes to aid in finding the points of interest) began to be used to excite
characteristic X-ray spectra (Hillier, 1947). With the addition of highly specialized
wavelength-dispersive spectrometers and after the introduction of raster beam scan-
ning, these initial instruments evolved into the first electron probe microanalysis
(EPMA) machines or the electron microprobes of today. The birth of these micro-
probe instruments would allow for the development of average atomic number, flu-
orescence, and adsorption correction factors (ZAF correction factors) that would also
be in use for later EDS and would pave the way for the development of SEMs in
analysis applications. These X-ray detectors specifically looked for the wavelengths
or peaks of specific chemical elements with specific detectors looking for specific el-
ements. They were used to determine the chemistry of analyzed points or spots and
exercised a precision that still has not been matched by auto-SEM energy-dispersive
X-ray detectors. In terms of terminology, the instrument is called an electron micro-
probe, the X-ray detectors perform wavelength-dispersive spectroscopy, and the
whole analysis is termed electron probe microanalysis. Duncumb (1999) gave a
detailed and anecdotal review of the history of EPMA. Rao etal. (2009) presented an
example of how microprobe data can be used to help characterize an iron ore.

5.2.3.2Automated mineralogical (auto-SEM) technologies


All current auto-SEM technologies are automated image analysis systems based on
SEMs as platform instruments with EDS detectors attached, the signals from which
are integrated according to the specific instrument. Furthermore, these EDS detec-
tors are usually run at low count totals to speed up analysis times and measurements.
Historically (pre-2008), up to four light-element EDS X-ray detectors could be at-
tached to any single system. However, since the analysis speeds have substantially in-
creased (analysis times have decreased), these numbers of EDS detectors are no longer
needed or used on most machines. All elements heavier than carbon can be identified
confidently although stoichiometric oxygen determinations can still prove problem-
atic, especially with low count spectra. For that reason, oxygen abundance is usually
based on the calculation of oxide mass percentages based on the analysis of the rele-
vant cations. Additionally, with the ubiquitous presence of carbon in both the mount-
ing media and carbon coating (which helps to ensure conducting specimens, which
in turn ensure consistent specimen currents and hence consistent BSE valueswith
calibration), carbon may seem more abundant than it actually is. All these instruments
have the ability to simultaneously utilize both BSE brightness and energy-dispersive
(EDS) X-ray spectra in the creation of digital mineral images. However, the rule sets
166 Iron Ore

(SIP files in the case of QEMSCAN, standard files in the case of MLA, and recipes
in the case of Mineralogic (R)) seldom account for both methods of mineral identifi-
cation due to historical paradigms. All these instrument paradigms have much to do
with the chamber/detector geometry and the traditional SEM setup of the developing
companies. Off-line data mining (which includes the off-line replaying or reanalysis
of datasets after mineral identification rules have been changed) and graph plotting are
possible with all the systems.
Differences in the values of quantities between auto-SEMs and other technologies
do occur, since the auto-SEM classification of minerals is based mostly on chemical
composition while traditional classification through XRD is based on crystal structure.
Also, whereas point-counting results (from light microscopes) are typically presented
in terms of area percent and XRD results are typically presented in terms of weight
percent, auto-SEM results are typically presented in terms of both area and weight
percent (calculated based on input density values since auto-SEMs cannot measure
density directly).
Since most auto-SEMs prefer flat surfaces for measurement (like light microscopes)
to avoid X-ray shadowing and the proper application of ZAF correction factors in the
EDS analysis, cognizance should be made of the fact that the morphological interpre-
tations of sections have limited value as evidenced in Figure5.2. Light microscopes
suffer from the same limitation. Historically, this is what has limited SFCA characteri-
zation on auto-SEM systems (though the same argument/limitation should be used for
light microscopes that have traditionally studied sinters).

5.2.3.3Auto-SEMs with specific regard to iron ore analysis


For all SEM-based systems, potential difficulties in analysis occur when phases of sim-
ilar chemistry have similar average atomic numbers (i.e., similar BSE intensities) mak-
ing distinctions difficult. While albite and quartz have similar average atomic numbers,
they are readily distinguishable due to their unique chemistries. Conversely, while py-
rite and pyrrhotite have similar chemistries, they are distinguishable due to their unique

3
1

Figure5.2 Morphological classification dependence on sectioning. (1) Original unsectioned


crystal. (2) Planar sectioning of crystal. The gray area illustrates a tabular interpretation of
sectioning. (3) Cross-sectioning of crystal. The gray area illustrates an acicular interpretation
of sectioning. Note how the same crystal can be classified differently according to its
morphology as determined by the obliquity of its sectioning.
Quantitative analysis of iron ore using SEM-based technologies167

average atomic numbers. Problem minerals therefore encountered in analysis may in-
clude hematite and magnetite since they share similar chemistries and similar average
atomic numbers. Goethite may also be problematic in identification, though it usually
contains sufficient impurities to help differentiate it from hematite, in auto-SEM sys-
tems. It is therefore important to note that auto-SEM distinctions between magnetite
and hematite are dependent on a number of factors, specifically the surface characteris-
tics of the iron oxides being measured (scratching, plucking, dipping beneath mounting
media, porosity, carbon coating thickness, etc.) and the degree of ionic substitution
being encountered in the iron oxides (in particular Mg, Ca, Al, and Mn).
Iron oxide discrimination on auto-SEM systems can furthermore be influenced by
BSE stabilizers (also known as BSE amplifiers) on certain systems, gun alignment
(essentially whether the electron beam is coming down the column unhindered with
no shadowing of the beam), and magnification field sizes. Furthermore, stage height
(working distance) focus primarily determines grayscale, while beam focus has an
effect on BSE grayscale averaging (and the image focus itself). Appropriate third point
BSE calibration standards and increased beam dwell times (i.e., giving the electron
beam more time on a specific point to make an identification) improve discrimina-
tion. Image analysis and processing can do away with certain artifacts of analysis.
Auto-SEM instruments make use of their own paradigms to reduce or eliminate some
of the abovementioned problems, though no single instrument eliminates all of them.
Figure5.3 presents images of an electron microprobe hematite standard measured in a
QEMSCAN system with different measurement parameters. The following sections look
more closely at the instrument types and varieties. Table5.2 highlights their differences.

5.2.3.4QEMSCAN
QEMSCANs were previously produced by Intellection on Carl Zeiss EVO platforms
(Pirrie etal., 2004; Gottlieb et al., 2000; Sutherland and Gottlieb, 1991). Currently,
QEMSCANs use FEI Quanta and field emission gun (FEG) platforms for their op-
eration. QEMSCAN instrumentation uses computer grayscale BSE imaging in con-
junction with a custom-designed X-ray spectral window database to identify phases
of interest. The QEMSCAN technique tends to produce X-ray maps, whereby EDS
X-ray spectra are collected at predefined beam-stepping intervals, at lower magnifi-
cations/higher field sizes (because of historically poor imaging). Because of its em-
phasis on X-ray collection, it is a preferred method for characterizing solid solutions
(where BSE grayscale is not amenable to species differentiation), finely intergrown,
chemically complex textures, and silicates, since EDS analyses are taken every couple
of microns (typically between 1 and 20 m). In recent years, QEMSCAN has focused
on collecting X-ray elemental concentration data (rather than data on X-ray spectral
windows) although QEMSCAN systems still favor superior X-ray collection.

5.2.3.5MLA
Previously manufactured by JKTech (the technology transfer company for the
Sustainable Minerals Institute housed in the Julius Kruttschnitt Mineral Research
Centre of the University of Queensland, Australia) on FEI Quanta platforms, MLAs
168 Iron Ore

(a) (b)

(c) (d)

(e) (f)

Figure5.3 Hematite standard measured in a QEMSCAN system with different measurement


parameters. Light gray=interpreted goethite; dark gray=interpreted magnetite; medium
gray=interpreted hematite. (a) Inappropriate third point calibration; (b) appropriate third point
calibration; (c) stage (1mm); (d) stage (+1mm); (e) focus (1mm); (f) focus (+1mm).

are now manufactured wholesale by FEI (Figueroa etal., 2011). MLA instrumen-
tation uses a best-fit model to compare collected EDS X-ray spectra with a spectral
library. The MLA technique tends to produce X-ray centroid measurements, whereby
single EDS X-ray spectra are collected for common BSE grayscale levels, at higher
magnifications/lower field sizes and is thus a preferred technique for small grain sizes
and high-resolution imaging. The MLA favors superior imaging over EDS collection.
In recent years, MLA data have been made compatible with QEMSCAN iDiscover/
iExplorer software, since it is more versatile for data interrogation and processing. A
novel feature of the MLA system is the ability to produce initial spectral collection
surveys where type spectra are collected for common BSE grayscales. The approach
Quantitative analysis of iron ore using SEM-based technologies169

Table5.2 Basic auto-SEM comparison among different platforms


QEMSCAN MLA Mineralogic TIMA
Historic SEM Oxford LEOs/ FEI Quanta N/A N/A
platform Carl Zeiss
EVOs
Current SEM FEI Quanta/ FEI Quanta/ Carl Zeiss TESCAN MIRA
platform FEG FEG EVO FEG/VEGA
EDX detection Oxford/Bruker Bruker SDD Bruker/Oxford PulseTor SDD
SDD SDD
Number of 4 1 N/A N/A
detectors
(historical)
Number of Up to 4 Up to 4 Up to 4 Up to 4
detectors
(current)
Traditional mode X-ray mapping X-ray N/A N/A
of measurement centroiding
Current mode of X-ray X-ray X-ray X-ray mapping
measurement mapping/X-ray centroiding/ centroiding/
centroiding X-ray X-ray mapping
mapping
Mineral ID Species Standard file Mineral recipe Mineral
identification matching identification classification
protocol rules/profiles
Accommodation 916 14 916 7 (Customizable)
of 30mm sample (Customizable)
blocks
Consistent Yes No (range No ?
specimen very narrow)
current for BSE
calibration
Spectral First match Best fit Priority/first ?
matching match (best fit
if priorities are
equal)
Iron oxide Arbitrary yet BSE Correlative ?
distinction consistent BSE histogram microscopy
cutoff peak
separation
Typical X-ray 1000 1000 >3000 ?
count rates (Variable)
170 Iron Ore

of the MLA system typically means that most systems use single-spectral libraries and
are also more beginner/user friendly than QEMSCANs.

5.2.3.6Mineralogic (R)
The Mineralogic (R) (and TIMA, discussed below) auto-SEMs are the newest au-
tomated mineral analyzers on the market. The Mineralogic (R) system is manufac-
tured by Carl Zeiss and favors image processing and image analysis that occurs during
the automeasurement to discriminate between the sample and mounting media (Hill,
2014). Unlike the QEMSCAN and MLA systems, analysis times are requested and
not analysis counts. The analysis involves a parallel two-stream approach where im-
age processing/image analysis forms one stream (and is in the control of the operator
unlike traditional auto-SEM technologies) and energy-dispersive X-ray analysis forms
a second stream (similar to more traditional technologies). The X-ray analysis and
image analysis can then be recombined to form a higher-order classification. This
may allow for mineral variety classifications in the future whereby, for instance, a
hematite spectral assignation occurring in a mineral grain of a specific shape might be
called specularite, theoretically. The Mineralogic (R) system also allows for a degree
of correlative microscopy whereby it is possible to georeference or correlate images
from different technologies (light microscopy, cathodoluminescence, secondary elec-
tron images, etc.) with those seen in the auto-SEM to facilitate mineral identification
(especially magnetite/hematite distinction).

5.2.3.7TESCAN integrated mineral analyzer


TESCAN integrated mineral analyzer (TIMA) is currently manufactured by TESCAN
(Kralova, 2013). Unlike traditional technologies, the TIMA has opted for PulseTor
SDDs. The relationship between TESCAN and PulseTor has allowed for an unprece-
dented degree of integration between the systems making X-ray/SEM communication
more seamless than traditional technologies. This in practice means fewer communi-
cation bugs and failures (crashes) compromising automated runs.

5.3Application of automated SEM-based


technologies to ore characterization
After an ore is measured but before it can be interrogated, most auto-SEMs typically
employ a step that entails rudimentary image processing. For the QEMSCAN and
MLA, this occurs after measurement and before reporting. For the Mineralogic (R)
system, however, this occurs during measurement but typically only after it has been
instructed to do so with an appropriate recipe.
The image processing, in all cases, entails one, more, or some of the following steps:
Touching particle processing (getting rid of or splitting touching particles).
Erroneous pixel processing (getting rid of slimes, unwanted artifacts, and dust contamina-
tion; in short, getting rid of any undesirables).
Quantitative analysis of iron ore using SEM-based technologies171

Mixed spectral resolution often referred to as boundary phase processing (getting rid of
or reclassifying mixed spectra occurring at boundaries between mineral grains of different
composition).
Granulating (through erosion/dilation algorithms; typically used to extract grains of eco-
nomic interest for further processing).
Field stitching (used for large particulates/slabs/cores that take up more than a single field of
view during measurement).

5.3.1Textural analysis
The state of the art for the latest auto-SEM technologies, in terms of textural analysis,
is mostly visual, although iDiscover (QEMSCAN software) does allow for catego-
rizing particles or grains on x- and y-axes creating what are termed image grids. The
properties that can be used to create those categories (things like shape factor) are,
however, somewhat limited. The MLA and Mineralogic (R) systems have far more
categories for interrogating shape. However, they lack the flexibility of being able to
create image grids. Textural images of conglomerate, measured with a QEMSCAN
system, are presented in Figure5.6.

5.3.2Mineral abundance
Table5.3 presents a typical example of how mineral abundance is presented in
auto-SEM analyses. Usually these data are presented in terms of mass percent. All
auto-SEM systems allow for first-order mineral identification (the SIP files, standard
files, and mineral identification recipes mentioned above) and first-order groupings
where mineral densities (and other mineral qualities) are assigned and second-order
groupings are simplified to make data interpretation easier and more relevant. So, in
terms of work flow, a QEMSCAN system may identify a zircon and a metamict
zircon from its SIP files, group these together in a primary mineral list (first-order
grouping), and further group these along with other minerals in an others grouping
(secondary mineral list) if there is not enough zircon to warrant specifying it in a min-
eral abundance table.

5.3.3Magnetite/hematite distinction
The analysis of iron oxides in automated mineralogy is primarily based on grayscale
values that are directly proportional to an average atomic number. It can thus occur
that, with atomic substitution (for instance, Mg or Ca for Fe), the average atomic num-
ber of magnetite can be lowered, resulting in a lower grayscale value, with a resultant
overestimation of hematite and an underestimation of magnetite. A pure hematite or
magnetite standard can be used as a third point calibration for both the QEMSCAN
and MLA systems although QEMSCAN typically uses copper as a third standard while
MLA uses chalcopyrite (Figueroa et al., 2011). Using a synthetically or artificially pure
standard might not negate these misidentifications always, since in naturally occur-
ring hematite, Fe can be substituted with Al, for instance, and in naturally occurring
172
Table5.3 A typical example of the bulk mineral abundance for conglomeratic iron ores
Chlorite/
Id Hematite Quartz Feldspar Mica clay Phosphates Pyrite Others Total
C1 40.9 57.0 0.0 0.2 1.1 0.4 0.0 0.3 100.0
C2 55.8 0.0 0.0 40.8 2.1 0.3 0.0 0.9 100.0
C3 74.1 20.2 0.1 1.5 2.3 1.4 0.0 0.4 100.0
C4 97.0 0.0 0.0 1.0 1.5 0.2 0.0 0.4 100.0
C5 33.0 0.0 0.0 0.2 63.2 0.1 0.0 3.4 100.0
C6 94.0 0.0 0.0 0.0 5.4 0.1 0.0 0.5 100.0
C7 78.5 0.0 0.0 15.9 1.5 0.3 0.0 3.8 100.0
C8 60.2 14.8 0.0 1.7 22.5 0.3 0.0 0.4 100.0

Iron Ore
Quantitative analysis of iron ore using SEM-based technologies173

magnetite, Fe can be substituted with Ca and Mg. Using a properly characterized standard
from a specific ore deposit may correct for this, bearing in mind that grayscales can
still be affected by mineralmineral or mineralbackground edge effects, particle relief
(minimized with appropriate polishing techniques), and porosity.

5.3.4Lithotyping/microlithotyping
The use of filters or categorizers and especially the creation of image grids (based on
whether particles are intuitively of a certain type) in iDiscover software allow for the
creation of rules that can separate particles into lithotypes, or rather microlithotypes.
Mineralogic (R) also allows for the very basic creation of expressions, usually the
addition of the quantity of two different minerals in an analysis, which are used to
classify particles during the measurement into defined microlithotypes. Since there is
no consistency between how QEMSCAN, MLA, and Mineralogic (R) (or any other of
the auto-SEMs) create these expressions, none of them can be moderated into standard
types that would correspond with geologic definitions of lithotypes. The software of
the various auto-SEMs also allows for very basic plotting of properties on ternary dia-
grams, with the three axes being limited to the particle properties that can be extracted
from the relevant software suites.

5.3.5Grain size
Since beam-stepping intervals (in the case of X-ray mapping) or pixel sizes (in the case
of X-ray centroiding) are always known and since this information can be counted for
contiguous classified pixels in mineral grains, grain size can be readily calculated
for mineral grains. Most particle/grain sizes are usually recalculated and reported as
equivalent circular diameter (ECD) values though this not necessarily always correct or
meaningful especially in the case of hematite bands in a banded iron formation (BIF).
An extracted hematite band converted to an ECD is meaningless. What must also
always be remembered is that the grain size calculated is very much dependent on the
magnification calibration of the individual instrument (and this is usually dependent
on the skill of the instrument service engineer).

5.3.6Liberation, locking, and association


Historically, mineral liberation has been used as a proxy for the processability of
a particular mineral, since the many properties of minerals that could allow for con-
centration and processing could not be directly measured from auto-SEM instrumen-
tation. Thus, the poor liberation of a mineral of interest in a tailings sample (and
consequent high liberation of a concentrate) was indicative of a feed that was properly
concentrated or processed. The high liberation of a tailings sample (and consequent
poor liberation of a concentrate) would be indicative of poor processing. High lib-
eration of a feed sample also suggested that it would be easily processed and poor
liberation of a feed sample would suggest difficulties in further processing. However,
liberation has been characterized differently by different mineralogists, companies,
174 Iron Ore

authors, etc., since different processing mechanisms are amenable to different mineral
properties and ore textures. Thus, a gravity or density separation procedure would ne-
cessitate liberation characterization based on the area percent of a mineral of interest
in a particle, whereas a leaching concentration procedure would necessitate liberation
characterization based on the free surface area percent (exposure on a particle surface)
of a mineral of interest in a particle. This latter, free surface area percent liberation
is called, by some authors, locking although the term locking is also applied to
minerals of interest sorted according to their association in an image grid with other
minerals, once all liberated mineral grains have been removed from an analysis. Here,
association is typically represented as the occurrence of binary particles where if a
particle constitutes more than 80% hematite and quartz, for instance, then it is classi-
fied as a hematite-quartz binary particle. These associations are subsequently quan-
tified. To avoid confusion, the proper use of the term association, it is felt, should be
restricted to the property of adjacency, where two minerals are associated, if they
are adjacent to one another. The number of occurrences of adjacency (as long as it is
specified as left-right pixel occurrences, top-down pixel occurrences, or both) can be
counted and converted into a percentage, and this would quantify an association be-
tween two minerals of interest.
Liberation is typically divided into categories based on 10% subdivisions of the
area percent occurrence of a mineral of interest (for instance, hematite) within a par-
ticle. Thus, particles are separated into 010%, 1020%, 2030%, etc., categories and
plotted on 3D charts with size fractions, liberation categories, and their weighting
(mass percent) being plotted on three axes, based firstly on unnormalized discrete
values (values that add up to the absolute amount of hematite in the total sample) to
normalized head (absolute) values (hematite is normalized to 100% for the total
sample) to normalized fraction (relative) values (hematite is normalized to 100%
per size fraction). These normalizations allow for the cumulative plotting of liber-
ation values on common graphs and allow for the direct comparison of liberation
between different samples, size fractions, etc. These liberation categories are usually
further simplified by grouping them in 20% categories. Thus, particles in the 020%
class are usually called locked, 2040% are lower-grade middlings, 4060% are
medium-grade middlings, 6080% are high-grade middlings, and 80100% are
liberated. Some workers suggest also using 0% (gangue) and 100% (fully lib-
erated) categories though since we are dealing with the 2D analysis of surfaces in
auto-SEMs, these categories are strictly speaking invalid since we cannot make
assumptions about the occurrence of hematite below (and eroded above) the sur-
face that has been analyzed.
As can be seen, the liberation characterization of minerals of interest in auto-SEM
systems is hardly trivial. Though with the direct input of mineral characteristics (such
as electrostatic affinity, floatability, and magnetic susceptibility) into auto-SEM an-
alytical programs and the understanding of particle characteristics (area percent liber-
ation vs. surface area percent liberation) that make minerals amenable to processing,
the reliance on liberation (as is currently conventionally used) may be reduced in the
future, with more of a direct reliance on the calculated processability of an ore.
Quantitative analysis of iron ore using SEM-based technologies175

5.4Characterization of natural and sintered iron ore


using QEMSCAN
The following applications of automated mineralogy in iron ore are based in
large part on the papers of Toneti and Dippenaar (2011) and Toneti etal.
(2012).
Toneti etal. (2012) used granulation on conglomeratic ores to extract hematite
clasts from matrix to obtain clast size distributions (Figure5.4). Image erosion (erod-
ing particle and grain edges one pixel at a time for a specified number of iterations)
was used to separate touching clasts and clasts from matrix in especially monomictic
(clasts of a singular mineral composition) and clast-supported (sensu stricto orthoc-
onglomerates) ores. Dilation was attempted to restore the clasts to their original size.
Granulation was further applied to sections of BIFs to facilitate the measurement of
band widths, as seen in Figures5.5 and 5.6.
A clast size distribution (quantified according to mass % hematite) was constructed
for these conglomeratic ores (Figure5.7). This allowed for a certain degree of ideal
grind size prediction to liberate hematite clasts from matrix (Toneti etal., 2012).
Plotting these clast sizes cumulatively allowed for the relative ranking of conglomer-
ates in terms of clast size (Figure5.8).
Similarly, hematite microbandwidth distributions were constructed using granula-
tion functions for BIFs, though there was no need for erosion functions to be used in
the image processing, to separate bands adequately. It was assumed that the bedding
bands were laterally continuous and thus that stereological corrections were not nec-
essary for accurate size determinations. An example of a microbandwidth distribution
is presented in Figure5.9.

Figure5.4 Image of originally measured slab and subsequent granulation of dark gray/black
hematite clasts with erosion and dilation operations having subsequently been applied.
White represents (vugular) macroporosity.
176 Iron Ore

3000.0 m

3000.0 m

Figure5.5 Hematite granulation applied to BIF samples. Note the interconnectivity across
hematite bands, which suggest the fracturing and remobilization of hematite.

Figure5.6 Example textures of conglomerates. Samples are approximately 9cm4cm.


Quantitative analysis of iron ore using SEM-based technologies177

80

70

60
Mass % hematite

50

40

30

20

10

0
>0 >250 >500 >750 >1000 >1250 >1500 >1750
Clast size (mm)

Figure5.7 Example of a clast size distribution histogram for a conglomeratic ore.

100

90

80
Cumulative mass % hematite

70

60

50

40

30

20

10

0
>0 >250 >500 >750 >1000 >1250 >1500 >1750 >2000 >2250 >2500 >2750 >3000 >3250 >3500
Clast size (mm)

C1 C2 C3 C4 C5 C6 C7 C8

Figure5.8 Cumulative clast size distributions for hematite in theoretical conglomerates.


Note prevalence of bimodal distributions indicating the occurrence of matrix versus clasts.
C4 (concave up, bottom right) would rank higher for clast size distribution than say C2
(concave down, top left).
178 Iron Ore

80

70

60
Mass % hematlte

50

40

30

20

10

0
>20 >40 >60 >80 >100 >120 >140 >160 >180 >200 >220 >240 >260 >280 >300 >320
Band width (mm)

Figure5.9 Example of a microband size distribution histogram for a BIF-type ore.

Hematite microbandwidths typically returned values on the micron scale, since


macrobands of hematite are typically contaminated with microlaminations of
quartz, which is to be expected when macroscopic hematite band transition to mac-
roscopic quartz bands. Put differently, microscopically, it was evident that there
were hematite microlayers present in quartz and vice versa for all the BIF sam-
ples analyzed. There was thus a disconnect between the size of hematite bands
seen in hand specimens (suggested by the geologist) and the small hematite bands
measured within the a uto-SEM. Hematite macrobandwidth compositions were thus
defined according to hematite mass percent thresholds and acted as a bridging con-
cept between microscopic hematite bandwidths and macroscopic hand specimen
bandwidths. Figure5.10 presents a hematite macroband profile as a function of
hematite composition for 500m intervals. Unfortunately, this complicated the
definition and delineation of hematite bands since one could define the occurrence
of hematite bands on the amount of hematite present within those bands. As such,
in Figure5.10, if one suggested that a hematite band must contain at least 50mass%
hematite (corresponding to ~35% Fe grade), then there are only two hematite bands
that can be accounted for in the sample even though there is clearly a third histo-
gram peak between the two larger hematite bands (and that can be seen in a textural
map of the sample in Figure5.11). This does however allow for minimum and
maximum bandwidths to be defined as outlined in Table5.4. Microbandwidth dis-
tributions have implications for product grade (quality). Macrobandwidth distribu-
tions, based on bandwidth compositions, have implications for tonnage (hematite
quantity).
The paper of Toneti and Dippenaar (2011) presents the first real attempt at trying
to solve the silicoferrite quantification problem on an automated SEM mineralogical
Quantitative analysis of iron ore using SEM-based technologies179

100

90

80

70

60
Mass %

50

40

30

20

10

0
500
1500
2500
3500
4500
5500
6500
7500
8500
9500
10,500
11,500
12,500
13,500
14,500
15,500
16,500
17,500
18,500
19,500
20,500
21,500
22,500
23,500
24,500
25,500
26,500
27,500
28,500
29,500
30,500
31,500
32,500
33,500
34,500
Stepping interval

Figure5.10 Demonstration of the subjectivity of defining band widths for sample B7 with the
occurrence of quartz microbands. Hematite composition is black while quartz is light gray.

system. Point-counting and XRD results provided for quantification baselines, bearing
in mind the inherent pitfalls that could be (and in some cases indeed were) found in
these types of abundance analyses. Table5.6 presents a basic comparison of mineral
distinction between QEMSCAN and point-counting techniques. Table5.7 presents
modal results of QEMSCAN against point-counting. Both these tables put in con-
text difficulties that can be encountered in both techniques when it comes to mineral
identification, for instance, Loo (1998) suggested that it is virtually impossible to dif-
ferentiate magnesioferrite from magnetite under optical microscope point-counting
conditions, whereas a magnesium X-ray signature is essential for QEMSCAN iden-
tification and is readily measured. Table5.8 presents a comparison of QEMSCAN
results with the Topas and Autoquan results (both XRD methods of measurement). In
Table5.8, an increase in magnetite is seen with an increase in magnesium (consistent
with Hsieh and Whiteman, 1993 and Oluwadare, 2007), with the increase being more
pronounced in the QEMSCAN results (as evidenced by the magnetite to hematite
ratio). High aluminum appears to promote SFCA formation as seen by the high SFCA
to calcium ferrite ratio (consistent with Oluwadare, 2007), with the results once again
being more prominent for the QEMSCAN analysis. The QEMSCAN results also sug-
gested the preferential formation or stabilization of hematite with an increase in alu-
minum (consistent with Oluwadare, 2007). Toneti and Dippenaar (2011) discussed
extensively possible reasons for the differences encountered in the SFCA modal
abundances between the various technologies but pointed to vague and inconsistent
180 Iron Ore

Hematite
Quartz

0 mm 5 mm 25 mm
Figure5.11 An example of a BIF measured within a QEMSCAN system. White represents
(vugular) macroporosity.

Table5.4 An example of bandwidth specifications for sample


above as defined by hematite mass % cutoffs
Hematite cutoff at 50% Hematite cutoff at 80%

Sample Min Max Min Max


B7 8mm 10.5mm 4mm 6.5mm

definitions of SFCAs in the literature and the dependence on morphological classi-


fication schemes resulting in a plethora of confusing SFCA synonyms. They pointed
out that chemical distinction schemes (which can be used by QEMSCAN) are sup-
ported by Scarlett etal. (2004) and Pimenta and Seshadri (2002). The paper also found
Al-deficient, pure SFCs and Mg analogs of SFCA. In summary, the paper showed a
rough agreement between QEMSCAN and XRD results, with less agreement between
QEMSCAN and point-counting results.
Quantitative analysis of iron ore using SEM-based technologies181
Table5.5 Potential auto-SEM measurement errors
(1) Sample preparation error (2) Measuring error (3) Data interpretation error
(1.1) Riffling/splitting/measuring/subsizing/ (2.1) Reproducibility (data) and repeatability (3.1) First-order grouping error
adequate sample error (instrument) (3.1.1) Density assignation error
(1.2) Density segregation/settling error (2.2) Statistical error (3.1.2) Mineral assay assigna-
(1.2.1) Associated with the addition of (2.2.1) Particle count error tion error
graphite (2.2.2) Measurement parameter error (line (3.2) Second-order grouping error
(1.2.2) Associated with density settling spacing, beam-stepping interval, field
in epoxy resin size/magnification errors, etc.)
(1.2.3) Bubble error (formation of (2.3) Block rotation error
bubbles in substrate) (2.4) X-ray peak overlap error
(1.3) Polishing/plucking/smearing/scratching (2.5) First-order mineral identification error
error
182 Iron Ore

Table5.6 A comparison of mineral distinction between


QEMSCAN and point-counting techniques
QEMSCAN (EDS-SEM Point-counting (reflective
Mineral microscopy) light microscopy)
Magnetite Fe and O signals above threshold Gray against hematite
grayscale value (85 89 depending
on deconvolution of BSE
histogram)
Magnesioferrite Mg, Fe, and O signals Mottled appearance in
magnetite of slightly
different gray (reflectance
is between hematite and
magnetite)
Calciomagnetite Fe and O signals with low-Ca Indistinguishable from
signatures magnesioferrite
Hematite Fe and O signals below threshold White against magnetite
grayscale value (85 89 depending
on deconvolution of BSE
histogram)
Calcium ferrite Fe and O signals with high-Ca Acicular morphology
signatures (dependent on sectioning)
SFCA Ca, Si, Fe, and O signals that may Tabular morphology
contain Al (dependent on sectioning)
Calcium silicates Ca, Si, and O signals (which may Only differentiable in
be differentiated according to polished thin section
Ca/Si)
Glass Everything else (though Everything else (when
differentiable into chemical distinguishable from calcium
subgroups) silicates)

All auto-SEMs allow for the input of standard mineral density values into their
analytic programs. It is therefore possible to calculate the average density of any
given measured particle by adding together the area fractions of minerals occurring
in a particle multiplied by their appropriate inputted densities. It is then possible to
place these minerals into density classes or categories and to plot a mass percentage
histogram. This is enough to gauge a rough idea of ideal heavy liquid separation
parameters for dense liquid separation, though by itself gives no idea of grade. If the
increasing cumulative percentage of iron is plotted along the same axes, along with
the decreasing cumulative percentage of any major gangue element (usually silicon),
the grade (and amount of deleterious element concentration) of a given concentrate
can be predicted if dense media separation is carried out at a specified dense liquid
cutoff. Conversely, if a specific grade of concentrate is required, it can be found on
the graph and the appropriate density of liquid needed can be selected (Richardson
and Morrison, 2003).
Quantitative analysis of iron ore using SEM-based technologies183
Table5.7 Comparison of QEMSCAN (QS) results against point-counting (PC) results (area %)
Sample 1 Sample 2 Sample 3 Sample 4

Mineral QS PC QS PC QS PC QS PC
Magnetite 6.3 12.8 3.2 7.8 9.4 7.0 5.8 8.1
Magnesioferrite 0.8 16.7 0.9 22.3 0.8 21.3 0.5 18.1
Calciomagnetite 13.0 0.0 11.5 0.0 12.5 0.0 12.4 0.0
Hematite 27.6 21.0 32.1 25.9 24.9 20.5 30.3 21.2
Calcium ferrite 30.4 11.3 29.3 6.4 30.7 6.2 33.0 14.9
SFCAa 18.4 35.2 17.4 32.7 16.5 40.8 14.8 35.9
Calcium silicates 0.4 0.0 1.0 0.0 0.3 0.0 0.1 0.0
Glass 3.1 3.0 4.6 4.9 4.9 4.2 3.1 1.8
Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0

a
Silicoferrite of calcium and aluminum.
184
Table5.8 Comparison of QEMSCAN (QS) results against XRD Topas and Autoquan (AQ) (weight %)
with additional chemistry
Name Sample 8 Sample 9 Sample 10
Aluminum 1.0 1.0 1.1
Magnesium 0.7 1.4 1.4

Mineral QS TP AQ QS TP AQ QS TP AQ
Magnetite 35.3 35.5 39.3 48.0 40.7 42.1 37.2 50.6 51.7
Hematite 22.2 34.7 33.3 11.2 24.9 25.2 25.5 24.3 24.6
Calcium ferrite 15.5 9.4 7.0 14.2 10.5 11.7 9.5 7.9 6.0
SFCA 20.1 12.6 12.8 19.1 12.8 11.5 18.5 10.0 11.0
Ca silicates/glass/ 6.9 7.8 7.6 7.5 11.1 9.5 9.3 7.2 6.7
others
Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0
SFCA/CFa 1.3 1.3 1.8 1.3 1.2 1.0 1.9 1.3 1.8
Mag/Hemb 1.6 1.0 1.2 4.3 1.6 1.7 1.5 2.1 2.1

CF, calcium ferrite.


a

Ratio of total magnetite to hematite (total magnetite=magnesioferrite+calciomagnetite+magnetite).


b

Iron Ore
Quantitative analysis of iron ore using SEM-based technologies185

5.5Summary
The relative strengths and weaknesses of automated SEM-based technologies for the
characterization of iron ore samples, compared with manual techniques, are summa-
rized below.

5.5.1Advantages (strengths)
While most auto-SEMs are not entirely operator-independent, they, to a large extent,
allow for the automatic collection of data once a measurement run has been initiated.
They are accurate once they have been taught what they need to know (though
this usually resides within the intellectual property of the operator or purchaser,
this does allow for a certain competitive advantage to be gained from purchasing an
auto-SEM and training it to do very specific measurements). Auto-SEMs are quick
with typical point analyses occurring in the region of milliseconds and measurement
runs per s ample block typically being resolved within 14h. There is no doubt that
these techniques are repeatable. Their consistency between testing runs is unparal-
leled. They remove the tedium of manual analyses often associated with microprobes
and more complicated instrumentation. While being expensive (in the order of 0.61.2
million US$), they are by no means the most expensive of analytic equipment and can
be cost-efficient since they can be run automatically after business hours and during
weekends. The fact that they can be left to run by themselves also means that they are
time-efficient in that operating mineralogists and metallurgists can be preoccupied
with other duties while their instruments are running.
The statistically valid data that are generated by these instruments run into the thou-
sands to millions of analysis points per measurement (approximately 10,000 times
greater number of analysis points than those obtained by point-counting) and most
can record and monitor variation during measurement to allow for project manage-
ment and quality control. The generation of these statistics means that work previously
considered impractical is now possible and that one is more likely to simply measure
for the sake of empirical experiment. Any single particle, collected and measured, can
be treated as a single particle or as part of a set of particles. As a single particle, its
properties are not hidden in a plethora of tables, modal results, and bulk data. These
auto-SEMs can, in a sense, be both holistic and atomistic.

5.5.2Disadvantages (weaknesses)
As has already been mentioned, for all SEM-based systems, potential difficulties in
analysis occur when phases of similar chemistry have similar average atomic numbers
(i.e., similar BSE intensities). These difficulties are minimized with appropriate stan-
dards, best practice methodologies including the stretching of contrast and the use of
multiple measures in combination (BSE, X-ray information, elemental ratios, etc.) to
increase confidence in analyses.
Further, there are a number of difficulties associated with automated im-
age analysis including the delineation of hematite clasts against hematite matrix
186 Iron Ore

compositions in these automated analysis systems. The occurrence of hematite


clasts in hematite matrices makes clast boundaries indistinguishable in auto-SEM
systems. Furthermore, if monomictic hematite clasts are clast-supported (sensu
stricto orthoconglomerates), then even if they are easily distinguished from a sili-
ceous matrix, they will still be touching one another and thus not easily be resolved
into separate clasts. The extensive image processing capabilities (specifically the
watershedding capabilities) of the Mineralogic (R) suite might allow for a superior
granulation that could allow for the separation of grains of the same minerals that
happen to be touching.
Furthermore, the sources and extent of error in auto-SEM analysis have yet to be
properly studied and characterized but can broadly be grouped into three categories
(not including error introduced during sampling and where the onus is not on the in-
strumentation to correct or quantify but on the samplers to be representative of their
sampling):
(1) Sample preparation error (such as inaccurate sizing/weighing, density settling, and incor-
rect polishing procedures).
(2) Measuring error (including data reproducibility and instrument repeatability).
(3) Data interpretation error (such as inappropriate mineral groupings and incorrect
densities)

Far be it from the scope of this chapter to elaborate on the possible causes of these
errors, for each of the above, the reader will briefly be made aware of where errors can
creep into an analysis by point form in Table5.5 (for more information, consult the
references at the end of this chapter and the author). Measurement errors are usually
less significant than sample preparation errors because, to a certain extent, they can
be compensated for in the mineral identification rule sets and can also, once again to
a certain extent, be corrected for in data processing by experienced operators. So, for
instance, boundary or edge effects can be accommodated for in mineral identification
rule sets by accounting for the presence of carbon (among other elements) from epoxy
resin and graphite fillers. Also, any unidentified (mixed spectral or boundary) phases
can be resolved into legitimate mineral entries by various data-processing strategies
(Toneti, 2014). More grave errors usually creep in with sample preparation (relief
artifacts, bubbles, plucking, smearing, scratching, etc.) and inexperienced auto-SEM
operators attempting to compensate too much for instrument physics (for instance, by
increasing mineral identification windows to limits that will catch multiple different
minerals within a single entry).

5.6Future trends
Doubtless, auto-SEMs will get faster, smarter, be more accurate, and be more au-
tomated in the future, the speculation of which lies in the how. Faster no doubt
lies in the improved performance of EDX spectrometers. In a single generation, liquid
nitrogen detectors were eclipsed by SDDs (silicon drift detectors), which were 4 times
quicker and doubtless will get faster in the future in terms of analysis speeds.
Quantitative analysis of iron ore using SEM-based technologies187

In terms of accuracy, detector quality is sure to improve and may in the future
approach the same level of elemental differentiability as WDS detectors though this
will probably require the writing of some very smart rules or code that operate in the
background of auto-SEM runs.
Smarter auto-SEMs will no doubt entail an improved query language especially
for textural evaluation, so that an operator mineralogist can bring up or return any
particles of interest that obey or observe any quality that has been measured directly
or calculated indirectly by inference. Many auto-SEM software programmers have
maintained (and doubtless will always maintain) that the data are all there It's just
a question of bringing it out. Forays into this area have already been made with most
auto-SEMs being able to at least accept additional data characteristics of minerals
other than the ubiquitous density (values such as flotation indices). Smarter query
languages or even smarter measurement variables may in the future allow for the dis-
tinction of mineral varieties based on their morphologies or textures. So, for instance,
specularitic varieties of hematite may be distinguished from martitic varieties of he-
matite based on shape factor (compactness, aspect ratios, etc.) or even their associ-
ation with pores (porosity). A smarter (or at least more integrative) approach in the
future may allow for the possible incorporation of cathodoluminescence (or any other
analysis method resulting in a multispectral approach) as a tool in grain segmenta-
tion or even further identification of minerals in iron ores (and other commodities).
Correlative microscopy that would combine SEM and optical approaches (and even
XRM) to overcome issues associated with the differentiation of phases with similar
chemistry and BSE values (magnetite and hematite) can most definitely add value to
iron ore quantification. This approach would not be far removed from the integrative
geographic information system (GIS) approach in common use for macro (large-
scale) geographic mapping.
More automated one-button approaches have already been developed by some
of the auto-SEM companies though great care and caution should be exercised with
these instruments to ensure valid datasets. As always quicker, less-intensive measure-
ment setup of samples (and subsequently less operator-intensive data processing and
interpretation) should allow the mineralogist more time to solve complicated problems
associated with mine, processing plant, or exploration surveys.
Some have speculated that X-ray microtomography may replace auto-SEMs in
the near future. This is unlikely. In all likelihood, XRMs will be treated in a comple-
mentary fashion with XRMs providing critical three-dimensional information to ores
through the mapping of attenuation coefficients, analogous to 2D BSE grayscales
(proportional to average atomic number), with these attenuation coefficients being
regularly compared, correlated, and validated with auto-SEM X-ray data. The main
impedances to XRMs completely replacing auto-SEMs (in the near future) are as
follows:
Their reliance on attenuation coefficients. One cannot directly gain access to X-ray informa-
tion within a core/sample. Just like auto-SEMs have overlapping BSE grayscales between
minerals, XRMs have overlapping attenuation coefficients. So only in very simplistic sam-
ples would it be possible to separate minerals in an ore through attenuation coefficients.
Once a suitable attenuation coefficient of a specific value has been identified, one would
188 Iron Ore

still have to figure out exactly what mineral is being looked at through nondestructive means
(commonly by using an auto-SEM).
Price. XRMs are currently far more expensive than auto-SEMs.
Memory. The data created by XRMs are currently prohibitively large for large-volume scans
(gigabytes of data produced at approximately tens of micron resolution for approximately
cm3 volumes).
Resolution. XRM resolution is generally low (also to do with the size of data created by an
analysisthe higher the resolution, the more prohibitive the memory usage).

There is however much scope for XRMs being used in a complementary fashion
to auto-SEMs.

References
Duncumb, P., 1999. Quantitative analysis with the electron microprobe. The first 50years
and beyond (Plenary Lecture). Fifteenth International Congress on X-Ray Optics and
Microanalysis (ICXOM), Antwerp, Belgium, August 2427, 1998. J. Anal. At. Spectrom.
14, 357366.
Figueroa, G., Moeller, K., Buhot, M., Gloy, G., Haberlah, D., 2011. Advanced discrimination
of hematite and magnetite by automated mineralogy. In: 10th International Congress for
Applied Mineralogy proceedings, Trondheim, pp. 197204.
Gottlieb, P., Wilkie, G., Sutherland, D., Ho-Tun, E., Suthers, S., Perera, K., Jenkins, B., Spencer,
S., Butcher, A., Rayner, J., 2000. Using quantitative electron microscopy for process min-
eralogy applications. J. Mineral. 52, 2425.
Hill, E., 2014. ZEISS Mineralogic MiningOre Process Optimization. Carl Zeiss White Paper.
Hillier, J., 1947. Electron probe analysis employing X-ray spectrography (Patent No. 2,418,029).
Hsieh, L.-H., Whiteman, J.A., 1993. Effect of raw material composition on the mineral phases
in a lime-fluxed sinter. ISIJ Int. 33, 462473.
Kralova, V., 2013. Automated mineral analysis in SEM for rapid evaluation of mineralogi-
cal samples. In: 51st Annual Conference of the Microscopy Society of Southern Africa
Technical Forum Proceedings, Pretoria, South Africa, pp. 25.
Loo, C.E., 1998. Some progress in understanding the science of iron ore sintering. In: ICSTI/
Ironmaking Conference Proceedings.
Oluwadare, G.O., 2007. Roles of alumina and magnesia on the formation of SFCA in iron ore
sinters. Trends Appl. Sci. Res. 2 (6), 483491.
Pimenta, H.P., Seshadri, V., 2002. Characterisation of structure of iron ore sinter and its be-
haviour during reduction at low temperatures. Ironmak. Steelmak. 29 (3), 169174.
Pirrie, D., Butcher, A.R., Power, M.R., Gottlieb, P., Miller, G.L., 2004. Rapid quantitative mineral
and phase analysis using automated scanning electron microscopy (QemSCAN); potential
applications in forensic geosciences. In: Forensic Geoscience: Principles, Techniques and
Applications. Geological Society, London, pp. 123136 (Special Publications No. 232).
Rao, D., Kumar, T., Rao, S., Prabhakar, S., Raju, G., 2009. Mineralogy and geochemistry of
a Low grade iron ore sample from Bellary-Hospet Sector, India and their implications on
beneficiation. J. Miner. Mater. Charact. Eng. 8 (2), 115132.
Richardson, J.M., Morrison, R.D., 2003. Metallurgical balances and efficiency. In: Fuerstenau,
M.C., Han, K.N. (Eds.), Principles of Mineral Processing. Society for Mining, Metallurgy
and Exploration (SME), Colorado, pp. 363365.
Quantitative analysis of iron ore using SEM-based technologies189

Scarlett, N.V.Y., Pownceby, M.I., Madsen, I.C., Christensen, A.N., 2004. Reaction sequences in
the formation of silico-ferrites of calcium and aluminium in iron ore sinter. Metall. Mater.
Trans. 35B, 929936.
Sutherland, D.N., Gottlieb, P., 1991. Application of automated quantitative mineralogy in min-
eral processing. Miner. Eng. 4 (711), 753762.
Toneti, I.., 2012. The characterization of surface overburden (pedocretes) at the Sishen
Mine, South Africa. In: FEI User Group 2012, Stellenbosch, South Africa.
Toneti, I.., 2014. Mixed spectral signature resolution with Zeiss Mineralogic Mining. In:
Carl Zeiss Lunch & Learn Workshop, Johannesburg, South Africa.
Toneti, I.., Dippenaar, A., 2011. An alternative to traditional iron-ore sinter phase classifica-
tion. Miner. Eng. 24 (12), 12581263.
Toneti, I.., Duncan, M., Bramdeo, S., 2012. The ore characterisation of different hematitic
iron ores. In: Process Mineralogy 2012 Proceedings, Cape Town, South Africa.

You might also like