Journal of Geosciences, 63 (2018), 47–63 DOI: 10.3190/jgeosci.

250

Original paper
Automated mineralogy and petrology – applications of TESCAN
Integrated Mineral Analyzer (TIMA)

Tomáš Hrstka1*, Paul Gottlieb2, Roman Skála1, Karel Breiter1, David Motl3
1
Institute of Geology, The Czech Academy of Sciences, v.v.i., Rozvojová 269, 165 00 Prague 6-Lysolaje, Czech Republic; hrstka@gli.cas.cz
2
TESCAN ORSAY HOLDING, a.s., Libušina třída 21, 623 00 Brno, Czech Republic
3
TESCAN Brno, s.r.o., Libušina třída 1, 623 00 Brno, Czech Republic
*
Corresponding author

The collection of representative modal mineralogy data as well as textural and chemical information on statistically
significant samples is becoming essential in many areas of Earth and material sciences. Automated Scanning Electron
Microscopy (ASEM) systems provide an ideal solution for such tasks. This paper presents the methods and techniques
used in the recently developed TESCAN Integrated Mineral Analyzer (TIMA-X) with Version 1.5 TIMA software. The
benefits from the use of a fully integrated quantitative energy-dispersive X-ray spectrometry (EDS) and an advanced
statistical approach to ASEM systems are demonstrated. Typically, the system can handle more than 500,000 X-ray
events per second. Using a common spectral total of 1000 events this represents the acquisition of 500 spectra per second.
A number of measurement modes is available to make the most effective use of these spectra depending on the application.
For a back-scattered electrons (BSE) map combined with EDS data with spatial resolution of 10 µm, this represents the
high-resolution measurement of c. 1 cm2 of a thin section or a polished rock surface in 30 minutes. A patented X-ray
spectrum clustering algorithm that lowers the chemical detection limit is described and an example of its use is shown.
The modal and textural (liberation, association, size etc.) data produced are statistically robust and provide information
across a broad range of Earth and material sciences. A comparison with some other available instruments is also provided
together with a number of case studies.

Keywords: TIMA, Automated SEM/EDS, applied mineralogy, modal analysis, artificial intelligence, neural networks
Received: 9 October 2017; accepted: 6 March 2018; handling editor: F. Laufek
The online version of this article (doi: 10.3190/jgeosci.250) contains supplementary electronic material.

1. Introduction instruments are still rather rare and have a limited use
because of the difficulties involved in automatically
Information on the modal composition of samples (i.e. identifying minerals by their color and other properties
volume % or weight % of minerals/phases present) and in reflected light and polarized transmitted light (Pirard
understanding of their textural relationships are essen- 2004; Donskoi et al. 2007; Lane et al. 2008; Poliakov and
tial in petrologic and petrogenetic studied (Le Bas and Donskoi 2014; Berrezueta et al. 2016).
Streckeisen 1991; Le Maitre 2002), modelling of ore Automated Mineralogy instruments such as QEM­
formation (Rollinson et al. 2011; Smythe et al. 2013; Nie SCAN (Gottlieb et al. 2000) and MLA (Fandrich et al.
and Peng 2014; Santoro et al. 2014) and in fact in any 2007) were initially designed mainly for use in mineral
research related to Earth or material sciences (Gottlieb processing to determine particle mineral liberation from
2008; Pirrie and Rollinson 2011). Optical microscopy, representative samples of plant products (i.e. feeds, con-
powder X-ray diffraction (PXRD), scanning electron centrates and tailings). These systems are indeed useful
microscopy (SEM), often supplemented with energy- for plant optimization and feed ore characterization for
dispersive X-ray spectroscopy (EDS) and electron-probe predicting concentrator and leaching performance from
microanalysis (EPMA) have represented the key meth- ore properties such as grade, elemental deportment, lock-
ods used to characterize material properties in the past. ing, grind size, hardness, and reagent consumption.
Although these techniques still play a crucial role in Lately, the AM systems have also progressively found
mineralogy, petrology and geology, the last two decades their way into many geological research laboratories
have seen the widespread use of SEM-based Automated where they are now being widely used for gathering
Mineralogy (AM) systems to help identifying minerals by mineralogical and petrological data. They have been em-
their composition and to quantify their proportions, size ployed to visualize and quantify petrological properties
and textural relations (Hoal et al. 2009; Pérez-Barnuevo such as alteration, contact zones, fractures, exsolution
et al. 2013). By contrast, automated optical microscopy structures, deformation re-crystallization and mapping

www.jgeosci.org
Tomáš Hrstka, Paul Gottlieb, Roman Skála, Karel Breiter, David Motl

distribution of inclusions as well as other data (Andersen also very important. In contrast to classical EDX phase
et al. 2009; Haberlah et al. 2010; Knappett et al. 2011; mapping software that requires high count numbers to
Rollinson et al. 2011; Santoro et al. 2015; Ackerman et be collected at each map pixel, the TIMA system is op-
al. 2017; Ward et al. 2018). timized to deal with rapidly acquired low-count spectra.
The AM systems have a number of time- and statistics- The TIMA has specific capabilities for geological
optimized analysis methods and modes of operation spe- applications because of its specialized integration of
cifically customized for various industrial and research hardware and software and its high level of automation.
tasks. These include particle scans, field scans, line scans, The speed, automation and unattended operation allow
specific mineral search routines and other more complex the researcher to collect detailed data that were not pos-
modes (Gottlieb et al. 2000; Fandrich et al. 2007). All sible to obtain through human-interactive operation of an
these AM systems are based on acquisition, presentation SEM/EDX system. The multiple EDX detectors, in addi-
and analysis of hyperspectral data (usually created by a tion to providing a speed multiplier, have a side-benefit of
combination of BSE and elemental maps) representing reducing shading for artefact-free measurement of uneven
a 2-dimensional (2D) image of flat mineral/phase sec- surfaces with an imperfect polish.
tions typically for large, statistically significant areas of Apart from the back-scattered electron (BSE), second-
the sample. A recent trend is that manufacturers of more ary electron (SE), and EDS detectors, the TIMA system
classic SEM/EDS hardware and elemental quantification can be also accompanied by cathodoluminescence (CL)
software are also starting to follow the example set by detector for simultaneous acquisition of CL data during
AM systems away from traditional (point analysis) and the measurements. An integrated Raman spectroscopy
“elemental mapping” to more useful “phase mapping”. system is also available as an option.
Their goal is to create 2D mineral/phase images from For the larger chamber, up to nine 27 × 47 mm thin
geological and other specimens though typically on a sections or fifteen 30 mm diameter or twenty two 25 mm
smaller scale due to the time constrains (Johnson et al. diameter polished blocks can be analyzed in a single
2015). It might be foreseen that the two approaches will batch. The smaller chamber accommodates two 27 × 47
possibly slowly converge while at the moment combina- mm thin sections or seven 30 mm diameter or 25 mm
tions of both concepts might be necessary to provide the diameter polished blocks. A fully automated system
right answers. (“autoloader”) that is capable of sample exchanges for
In this article we summarize some basic concepts of one hundred 30 mm diameter or 25 mm diameter polished
the use of AM systems in geosciences. Moreover, we blocks is also available.
demonstrate the latest developments of the TESCAN All the processing of spectral phase classification is
Integrated Mineral Analyzer (TIMA-X) system and its performed already during the data collection in order to
applications on several typical examples. Some sugges- use the time efficiently and to produce results by the time
tions are also made for further developments towards when the acquisition is finished. The raw data are also
universal, statistically sound instrumental methods for saved together with the processed results to allow for
the analysis of minerals and rocks. modification of the parameters and off-line reclassifica-
tion and/or reinterpretation of the measurements.

2. TESCAN Integrated Mineral Analyzer

(TIMA) 2.2. Back-scattered electron (BSE) image
analysis
2.1. TIMA hardware Imaging and image analysis are fundamental to automated
mineral analysis. A stable and well calibrated BSE image
The TIMA-X system is based on one to four Energy is a prerequisite for its proper implementation (Fandrich et
Dispersive X-ray silicon drift detectors (SDD) attached al. 2007). The machine calibration is performed on a Fara-
to standard LM (small) or GM (large) chambers of the day cup and a platinum standard; automated checks and
TESCAN MIRA (field-emission gun – FEG) or VEGA optimizations for long-term measurements are included in
(thermionic emission – tungsten) platforms. the system. At the typical working conditions, the stability
High throughput is achieved by the use of multiple of the probe current is better than 0.5 % in 24 hours (for
TIMA EDS detectors operated at very high count rates (in both tungsten and field-emission systems). Consequently,
excess of 2,000,000 counts per second, as measured on a the BSE signal, which depends on the probe current, dis-
platinum standard) and using low-count EDX spectra to plays the same (0.5 %) level of stability. It means that if
identify the minerals at each measurement point (or each a typical mineral has a mean BSE level of 50.00 arbitrary
homogeneous segment/part of a grain depending on the units, it will not change more than ± 0.25 in 24 hours. The
analysis mode). Hardware and software integration is probe current and BSE detector parameters are adjusted

48
 Automated mineralogy and petrology – applications of TESCAN Integrated Mineral Analyzer (TIMA)

Mean atomic number

13.0
12.8
12.6
12.4
12.2
12.0
11.8
11.6

muscovite

pyrope
dolomite
labradorite
microcline

tschermakite
tremolite

bytownite
clinochlore

heulandite-Ca
anthophyllite

leucite
magnesioaxinite

zoisite
chrysolite
gedrite
prehnite

staurolite
actinolite

ferroaxinite

ankerite

diopside

pargasite
orthoclase

anorthite
adularia

edenite
enstatite
spinel

gypsum

winchite

clinozoisite
glaucophane

phlogopite
sanidine

magnesiohornblende
Fig. 1 Mean atomic number (Z) plotted for selected minerals and phases to demonstrate the BSE overlaps.

automatically to maintain optimum stability. This also al- 2.2.1. Particle de-agglomeration
lows for easy discrimination of phases on a long-term run
with BSE image only. For particulate samples such as mineral sand, dust,
Due to a number of factors the BSE calibration curve soils or milled products the TIMA system has an auto-
of individual BSE detectors is not linear, which creates mated touching-particle function that detects agglomer-
potential problems in cross-correlation with other sys- ates (Hrstka 2008; Kwitko-Ribeiro 2012). The TIMA
tems. Therefore the use of the average atomic number (Z) separates particles according to a set of predetermined
to monitor the BSE response of a mineral/phase could be parameters. This functionality is crucial if textural char-
suggested to overcome this calibration problem. Lloyd acteristics of individual particles/grains are of interest to
(1987) has examined in detail the use of BSE signal in prevent false association or size-distribution data.
mineralogy.
It should be noted that the short-term stability and
2.3. Phase segmentation
resolution are generally much higher than the long-term
one (~0.1 Z compared to ~1 Z or more). This leads to the Once individual segments have been identified by the
important ability to use the relative BSE brightness for initial BSE image, the next step of the mineralogical
image/phase segmentation, but gives it a more limited use analysis is to delineate coherent segments of BSE and
in the actual mineral/phase identification. EDS spatial data by the clustering/edge detection method
There have been attempts to use the BSE-only ap- (Motl and Filip 2013). The segmentation is an impor-
proach (Harding 2002; Sánchez et al. 2012) but its tant step that enables the creation of high-count spectra
application to more complex tasks is rather limited due suitable for EDS quantitative analysis. This unique ap-
to the common overlaps of BSE values of individual proach provides a means by which to assign the directly
minerals and their high variability even within a single measured chemistry to each of the measured phases as
mineral species (Fig. 1). Among the ~6,000 known needed. The individual low-count spectra typically used
minerals and valid unnamed minerals (Smith and Nickel in AM systems (~1000 to below ~20,000 counts) are not
2008), there are over 700 cases of overlapping BSE suitable for generating precise and accurate quantitative
image data because of differences in average atomic elemental information.
numbers being below 0.1 Z. This does not even take into The segmentation is done after data acquisition, by
account variations based on difference in BSE within a creating a gradient image using the BSE image and the
single mineral species due to elemental substitutions. elemental X-ray intensities. The gradient image is then
Therefore the use of combined BSE and EDS is strongly processed by a watershed transformation that detects
advisable. continuous edges. The watershed transformation uses

49
Tomáš Hrstka, Paul Gottlieb, Roman Skála, Karel Breiter, David Motl

High resolution mapping a) Point spectroscopy b)

10 K PARTICLES ~ 168 min (5 040 000 analytical points) 10 K PARTICLES ~ 2 min (40 000 analytical points)

BSE 2,5 µm & EDS 2,5 µm step BSE 2,5 µm & Single EDS per segment/grain

Line mapping c) Dot mapping d)

10 K PARTICLES ~ 22 min (660 000 analytical points) 10 K PARTICLES ~ 43 min (1 280 000 analytical points)

BSE & EDS 2,5 µm step on a line - lines 30 µm apart BSE 2,5 µm & EDS 5 µm step + Single EDS for small grains

20 µm

EDS ANALYTICAL POINT BSE GRID GREATEST INSCRIBED CIRCLE

Quartz Feldspar Topaz Zircon

50
 Automated mineralogy and petrology – applications of TESCAN Integrated Mineral Analyzer (TIMA)

one parameter – the minimum gradient that is regarded 2.4.2. TIMA Point Spectrometry (TPS)
as an edge. The value of this parameter is controlled
by the user. Whenever two segments in contact have In point spectrometry, individual particles and distinct
a similar composition and, as a result, the gradient phases/grains are determined using the BSE image only.
between them is low, they are merged into a single Areas of similar BSE brightness are identified as homoge-
segment rather than remain separated. The minimum neous regions. The center of the largest inscribed circle of
gradient parameter is used to increase the separation of each region is then used as a single X-ray analysis point
such phases. If it is too high, the software creates too to identify a mineral or a phase representing the entire
many segments and, as a result, this yields large datasets region (Fig. 2b). This method is very fast, but cannot
and off-line reclassification and data processing takes resolve two adjacent phases with similar mean atomic
more time. numbers but different chemistry and thus can lead to
large areas of misclassification (Fandrich et al. 2007). It
can be used to handle various geological samples as far
2.4. Mineral identification by BSE and X-ray as the minerals of interest have sufficient BSE grey level
analysis (Acquisition Modes) contrast. This can be difficult to predict for unknown
The TIMA currently has four X-ray analysis scanning samples and thus only well-known/previously studied
modes to identify mineral species: High-resolution samples can be recommended for this type of analysis.
mapping, Point spectrometry, Line mapping and Dot Its definite advantage is the high speed as only a single
mapping. These modes have been developed to econo- spectrum is collected per homogeneous BSE segment.
mize on the X-ray acquisition time. Each mode can be As stated previously, Z resolution is highly dependent on
further optimized to perform specific tasks. The most measurement parameters like beam current, brightness/
important optimization parameters are the number of contrast settings of the instrument, quality of the sample
X-ray counts per analytical spectrum and the distance surface, focus etc. (Lloyd 1987; Harding 2002).
spread (pixel spacing) of individual analytical points
(Fig. 2).
2.4.3. TIMA Line Mapping (TLM)
Line mapping is used to reduce X-ray acquisition time
2.4.1. TIMA High-resolution Mapping (THRM)
but still maintain some textural information such as
In this mode the TIMA first collects the BSE signal at the mineral grain size and mineral association. For line map-
analytical point and then one X-ray spectrum (typically ping each field is covered by equidistant horizontal lines
> 1000 counts) for each point over a regular grid within using a specified “line spacing” (Fig. 2c). The electron
each segmented particle. The BSE level is used as a beam moves along each line in regularly distributed
threshold so that if the BSE level is outside the specified measurement points using the specified “pixels spacing”.
range (e.g. epoxy), the X-ray spectra are not collected in At each point, the BSE level is determined, if the BSE
order to save run time (Fig. 2a). This method can be used level is above the threshold, the beam is kept on this spot
to handle various rocks with complex textures or other until the specific number of X-ray counts from the spec-
samples where detailed resolution is required. Due to the trometer is collected. The beam then moves to the next
optimized data processing and simultaneous use of four point. When the pre-determined limit of the line length
SDD detectors, the TIMA acquisition rate, when trans- (field size) is reached, the beam moves to the next line
lated into measurement time elapsed, is approximately and starts over. This procedure is repeated until all lines
30 minutes for the analysis of 1 cm² of a granitic rock at covering the selected field have been scanned.
10 µm pixel spacing and 1000 counts per spectrum per Then, the individual lines are divided into multiple
point (in the THRM mode). This acquisition mode, to- linear sections using the combination of the BSE level
gether with the liberation analysis, can be used to collect and the EDS data to determine boundaries between dis-
modal and textural data (e.g., Žák et al. 2016). tinct phases.
The EDS data from the measurement points within
each line section/segment are summed. The mean BSE
level for each line section/segment is determined as well.

Fig. 2 Sketch illustrating the principles of TIMA acquisition modes. The mean BSE level and the combined spectrum are used
a – High-resolution mapping using the same regular BSE and EDS to classify the line sections using the specified classifica-
analysis grid. b – Point mapping uses BSE only to segment the image tion scheme to determine the mineral/phase.
and collect single EDS spectra from each phase discriminated based on The advantage of this method in comparison with
BSE. c – Line mapping records one-dimensional profiles through the
Point Spectrometry method is that it resolves boundaries
sample. d – Dot mapping uses different pixel spacing for the BSE and
EDS data collection. between phases with similar mean atomic number but

51
Tomáš Hrstka, Paul Gottlieb, Roman Skála, Karel Breiter, David Motl

different chemistry, and still provides high speed. This

method gives only bulk association and modal data due to
one dimensional cuts through the particles rather than two
dimensional maps of each particle. However, it has the
advantage of enabling the measurement of more particles
to obtain better sampling statistics.
It can be used to handle various complex geological
samples with the advantage of high speed if the collection
of the full set of quantitative textural data is not critical. It
also stores BSE images of each field so the operator can
assess the quality of the results (perform QA, i.e. quality
assurance) easily and get detailed qualitative information
on textures from the BSE images (Fig. 3).

2.4.4. TIMA Dot Mapping (TDM)

Fig. 3 Example of thin section modal analysis by the means of TIMA Dot mapping imposes a BSE grid with specific resolution
Line Mapping (TLM). Quantitative modal data are obtained fast (10 min (determined as “pixel spacing”) over the entire sample,
for the whole thin section) while textural context is provided by the or specific particles/grains thereof (based on the BSE
simultaneously acquired BSE image.
threshold). It uses the acquired image to segment areas
of homogeneous BSE intensities and identifies the center

a)

e)
d)

c)
b)

Fig. 4 Examples of a basic classification scheme with classification rules and mineral data for quartz selected. a – Database/list of minerals in a
basic classification scheme. b – Standard spectra taken from the measured sample or calculated from stoichiometry/theoretical values. c – Chemical
data acquired from the EDS during the measurement or imported from an external source. d – Classification rules automatically generated based
on the standard spectra (b). e – Mineral properties like name and density assigned to the selected mineral entry.

52
 Automated mineralogy and petrology – applications of TESCAN Integrated Mineral Analyzer (TIMA)

of greatest inscribed circle. In the second pass it creates a tions. The phase chemistry for a specific classification
rectangular grid for X-ray acquisition with resolution spec- scheme can be based directly on the collected EDS data,
ified by the “dot spacing” over each of the “preliminary” taken from the theoretical stoichiometric composition, or
BSE phase segments determined in the first step. The com- entered manually based on a relevant external analysis
bination of the high-resolution BSE image and the lower- [e.g. EMPA, Laser-Ablation Inductively Coupled Plasma
resolution EDS data is used to greatly improve the phase Mass Spectrometry (LA-ICP-MS), Proton-Induced X-ray
segmentation when the preliminary segments comprise Emission (PIXE), or Micro X-ray fluorescence (µXRF)].
of multiple minerals/phases with similar Z but different These external analyses can account for elements which
chemistry. X-ray data from zones of similar BSE and are sometimes present in phases in trace amounts but are
EDS signals are summed to produce single higher quality important for the overall element distribution of the sample
spectra for each final segment. Average BSE and summed (e.g. gold in pyrite, nickel in goethite). This is particularly
spectra from each final segment are used to determine the important for phases containing elements with concentra-
mineral identity. The origin of the second X-ray data grid tions below EDS detection limits.
is centered on each segment. This means that even grains The density as a first approximation is based on the
smaller than the “dot spacing” of the X-ray data grid are mineral/phase identity or external study of the mineral/
characterized by at least a single X-ray analysis point. phase. There is no direct means currently available for
The Dot Mapping method provides an excellent measuring mineral density inside a SEM.
compromise between the high-resolution mapping and Built-in QA routines provide a confusion matrix to
the point spectrometry method in terms of speed versus point the user to potential problematic phases which
textural/information detail. As a single X-ray spectrum is could be misidentified/mismatched due to overlaps in the
collected even for particles smaller than the “dot spac- classification parameters. Basic classification schemes are
ing”, even the smallest bright particles and textures are provided with the system as a starter kit. The Webmineral
recorded (Fig. 2d). database (Mineralogical database 2017) is included to al-
The data collection procedure can be customized for low for quick and easy creation of classification scheme
specific tasks by the variation of its parameters (e.g. very entries from theoretical mineral formulae matching the
fast analysis of a thin section – Slavík et al. 2016; Acker- specific project needs.
man et al. 2017) or very detailed, submicron analysis of In order to further simplify the current workflow, a
complex dust particles or soil contaminants (Harvey et supervised or fully autonomous system based on artificial
al. 2017; Hrstka et al. 2017a, b). intelligence, machine learning and neural networks is
under development (Hrstka et al. 2017a).
2.5. Mineral standards library (Classification
Scheme) 2.6. Measurement analysis types
Mineral identification through BSE signals, EDS data The TIMA software uses several acquisition modes (Sec-
or the combination of both requires a library of mineral tion 2.4 above) to create various Measurement Analysis
compositional limits/mineral definition rules or standard Types designed to accommodate different mineralogical
spectra (Classification scheme in TIMA – Fig. 4). Any and petrological information requirements. The Measure-
unknown spectra collected from a point or a segment, ment Analysis Types and their associated Acquisition
together with related information from other detectors Modes are summarized in Tab. 1 and described below.
(BSE, CL, SE etc.), are used to find the closest or first
match to the records in the database providing mineral/
phase identification. For TIMA, a first match principle is 2.6.1. TIMA Modal Analysis (TMA)
used being based on the combination of mineral definition This is the most basic analysis method in which only the
rules, either determined automatically (from a “standard modal data are collected through “high-resolution map-
spectrum” or calculated from a theoretical mineral/phase
composition) by the system or created by the user from Tab. 1 Available analysis types and relevant acquisition modes
BSE, X-ray spectral windows counts and/or their ratios.
The automation in creating the rules based on standard Analysis type Modal Liberation Bright phase Section
spectra significantly reduces the time spent on building a Acquisition mode analysis analysis search analysis
High-resolution
classification scheme, especially when tens or hundreds × × × ×
mapping
of potential phases are involved. A classification scheme Point spectroscopy × × ×
provides not only the name of the identified phase, but also Line mapping × ×
information on its exact chemistry and density needed in Dot mapping × × ×
all subsequent modal or elemental deportment calcula- × = Currently available

53
Tomáš Hrstka, Paul Gottlieb, Roman Skála, Karel Breiter, David Motl

ping” (see Section 2.4.1). It can be applied to particulate user to develop more advanced algorithms for phase
material or thin sections and polished blocks to obtain segmentation and clustering even for the most complex
statistically robust modal data. It is in a way similar to samples (Hrstka et al. 2017a, b).
classical point counting (Glagolev 1934; Larrea et al.
2014 and references therein), but is able to collect more
than 500 analytical points per second! 2.6.3. TIMA Bright Phase Search (TBPS)
This mode is a specific instance of the liberation analysis
2.6.2. TIMA Liberation Analysis (TLA) that is optimized to search for phases (usually scarce)
with a specific BSE signal, chemical composition, or
In the Liberation analysis (Fig. 5), the EDS data are col- both. It adds an extra step of initial assessment of the
lected simultaneously with the BSE or in the subsequent phase as a phase of interest either based on a BSE value,
phase depending on the acquisition mode. In general, or on a single spectrum, i.e. phase identification based on
the Liberation analysis can be customized to provide a a certain ”classification scheme” (phase filtering). This
very versatile tool for collecting statistically robust data provides optimum results for search for specific phases
on modal and textural characteristics of the sample. The of interest as demonstrated in some previous studies
patented segmentation algorithm (Section 2.3 above) (Haluzová et al. 2015; Slavík et al. 2016). There is also
combines the BSE and X-ray spectral information from a parameter to decide whether the entire host particle
similar points in a grid automatically into clusters (zones should be subjected to EDS analysis or if only a certain
of coherent BSE and EDS data). The EDS data from area around the grain of interest is analyzed (Fig. 6). This
the measurement points inside each grain/segment are provides the means of analyzing specific grains within
summed to produce high-quality spectra. The mean BSE thin sections or large particles and include the relevant
level for each grain/segment and the combined spectrum textural/association/locking data. It can dramatically
are used to classify the segments and form grains using speed up the analysis in such scenarios. If required,
the specified classification scheme to determine the min- BSE images of the whole sample can be also collected
eral/phase identity. The raw unsegmented data are saved simultaneously in order to understand the general textural
and can be re-segmented later off-line. This enables the context.

2.6.4. TIMA Section

a) b)     Analysis (TSA)
This analysis type detects phases
using either BSE or a combina-
tion of BSE and EDS data. The
EDS data are collected simulta-
neously with the BSE or in the
subsequent phase depending on
the acquisition mode. This mode
does not detect the particles; it
BSE EDS is designed solely to study pol-
ished sections or thin sections of
c) d) a solid rock.

Fig. 5 Individual steps of TIMA Lib-

Segments

eration analysis with dot mapping

mode – BSE image at 0.3 μm (a) and
particle

EDS data at 1.5 μm (b) are combined

by an algorithm to produce pools of
coherent BSE and EDS data/segments
(c). Energy-dispersive data from each
pixel within a segment are summed
to produce one high-quality spectrum
grains (suitable for quantitative EDS analy-
sis). The average BSE intensity is also
Segmentation Mineral Map calculated. Data from each segment are
then compared against the classification
50µm scheme and mineral map is created (d).

54
 Automated mineralogy and petrology – applications of TESCAN Integrated Mineral Analyzer (TIMA)

2.7. Data analysis and presentation tabase structure, to view them interactively, to process
them, analyze and create reports to present the quantita-
In addition to the measurement capabilities, the TIMA tive mineralogical and petrological information generated
software enables saving the measurements/data in a da- by the TIMA measurements. Image data (currently BSE,

a)
b)

c)

d)

Fig. 6 Specific mineral search in thin sections from the Ni–Cu–(PGE) mineralization at Rožany, Czech Republic. a – Example of a thin section
covered by 904 individual fields with 600 μm field width (resolution of 0.5 μm, 5000 counts per point and 5 μm neighborhood size). This measure-
ment identified ~2000 potential targets on a single thin section and automatically identified 43 PGM grains. Summed EDS spectra provided their
composition and exact coordinates for navigation of the consequent EMPA analysis. b – Zoomed area with three PGM grains with their surrounding
minerals being analyzed to provide association/locking data (e.g. presence in sulfides, silicates or on the grain boundaries). c – Selected particles
of PGM with information on their area, size and position within the sample. d – Statistical data on the number of individual minerals analyzed
during the automated search and the area (volume of phases in pixels) of each mineral group.

55
Tomáš Hrstka, Paul Gottlieb, Roman Skála, Karel Breiter, David Motl

SE and CL) are combined with the elemental composition 2.8. User-defined expressions and queries
and densities of the identified phases/minerals on a pixel
by pixel (or segment by segment) basis to produce a va- User-defined expressions are used for particle classifica-
riety of mineralogical data. There are several predefined tion, filtering and sorting. This provides the best opportu-
reports/data analysis tools that are available; the informa- nity to explore the data for any specific research project
tion they provide is nearly self-explanatory: and create outputs based on categorizers. Examples
● Panorama, Fields, Particle viewer, and Grain view could be plotting drilling log data from several samples
are used to visualize the images of grains, particles, studied during a drilling campaign, geological mapping,
individual analyzed fields or whole samples. provenance study, creating complex tables for dashboards
● Mineral association, Mineral liberation, Mineral and interactive real-time work with data.
locking, Mineral release, Grain size Phase SSA,
Porosity and Particle constituents are designed 2.8.1. Expressions and queries
to provide values related to textural analysis of
the samples. For the examples of textural analysis The TIMA software has the ability to select multiple sub-
reader is referred to e.g. Lastra (2007) or Pérez- sets of measured data from the database by user-defined
-Barnuevo et al. (2013). queries. The syntax is somewhat similar to a Structured
● Elemental Mass, Elemental deportment and Mineral Query Language (SQL) approach but is specifically tai-
Mass provide results on modal distribution of mine- lored to AM requirements.
rals/phases and the distribution of elements between Queries can be used to extend capabilities of the exist-
individual phases/minerals. The performance of a ing predefined reports or to create new specific custom
theoretical perfect concentrator is given by the Gra- reports. The query can be understood as a function that
de-Recovery report. It is used to compare with the derives a statistics from a (sub-) population of objects, for
actual performance of a mineral concentrator plant example total mass of a phase. Depending on the report
and asses its efficiency. See Lotter et al. (2011) and type, the statistic assessment is applied to all selected
Altree-Williams et al. (2015) for more details on the measurements or a subpopulation thereof classified using
use of grade-recovery data. the categorizer or created by filtering of a certain type of
An important type of report is the Assay reconcilia- objects. This is a highly customizable way to explore the
tion that provides a QA check against an independent data and to provide appropriate interpretation suitable for
chemical assay of a sample. If a representative sample every specific mineralogy/petrology research or industry
is selected, then two aliquots can be made. The first driven task. A suitable example could be looking at the
is measured by TIMA and the modal mineralogy is deportment/distribution of a certain element among dif-
combined with the mineral chemistries and densities to ferent mineral phases or even individual lithotypes pres-
calculate a theoretical elemental assay of the sample. ent in the sample.
The calculated assay is compared to the XRF, ICP-MS
or any other independent bulk chemical analysis of the 2.8.2. Filtering
other representative aliquot. Very good reconciliation
can be achieved but some potential pitfalls related Filters are used to create an object (e.g. particle or min-
mainly to the sampling statistics and the quality of eral grain) sub-set for subsequent inclusion in the com-
the preparation of the samples were identified and putation and display of images, tables and charts. A filter
discussed in literature (Hrstka 2008, 2012; Johnson et can be as simple as manually including or excluding a
al. 2015; Lastra and Paktunc 2016; Pooler and Dold sub-set of measurements or it can be more complex,
2017). based on user-defined criteria.
Measurement properties report tabulates measure- Multiple object properties can be combined in a filter
ment parameters such as the measurement type, acqui- through expressions or through a selection of multiple
sition mode, pixel spacing with operational statistical filters from a pre-defined table. Multiple filters are a form
values such as measurement time, area, number of of the logical AND operator; an object must pass all the
fields, number of particles and number of analytical conditions in a filter group (referred to as “expression”)
points acquired. to be included.
There are also general-purpose reports, which include An example of application is the location of grains of
the Generic output, Particle category and Category view- zircons in a heavy minerals grain mount for later analysis
er. Each of them can be customized for a particular pur- by e.g. EPMA or LA-ICP-MS. A multiple filter can eas-
pose by using mineral particle and elemental properties ily be defined to show particles with zircon grains larger
from the measured data in expressions and queries. These than 30 µm (e.g. combined size (size>30) and phase
reports represent the full potential of the AM systems. (MineralPercent(Zircon)>90). It can also be

56
 Automated mineralogy and petrology – applications of TESCAN Integrated Mineral Analyzer (TIMA)

used to exclude some artefacts from the measurements combined modal and textural criteria at the same time
based on their size, composition and/or shape. Filtering for categorizing, this approach of automated rock type
is also very useful for interactive work where it can be classification can be also adopted for sedimentary and, to
used to focus onto a small subset of relevant information a certain degree, even to metamorphic rocks.
(e.g., particles with density > 3 g/cm3 and size > 100 µm).
2.8.5. Mineral grouping
2.8.3. Sorting
As the list of phases/minerals identified in a sample can
Sorting allows the user to arrange objects (e.g. particles be quite extensive, a grouping function allows a mineral
or grains) based on their properties. It allows the user to group to be defined and further treated as individual
visualize particles based on their size, mineral or element mineral phase (e.g. all feldspars, all amphiboles, all sili-
contents, density or other parameters. This is particularly cates, all heavy minerals etc.). Minor minerals of little
useful while performing the quality control on the results importance to an analysis could be automatically grouped
or for understanding data present in calculations visually. as “other” based on a minimal mass in the sample thresh-
An example would be to sort all the zircon grains found in old. Several groupings with different level of detail for
a thin section by their size or sort all the grains found based classification of minerals/phases (e.g. species, series,
on their density to evaluate the phases likely allowing the subgroups, groups) can be used at the same time to
LA-ICP-MS dating. Another example would be to sort all simplify the navigation through data or results reporting
the particles/grains by element mass, or phase mass to see (e.g. detailed grouping for modal analysis and simplified
the most important carriers of a certain element/phase in the grouping for textural/petrographic analysis).
tested sample (e.g. to evaluate the possible nugget effect of
one big grain hosting most of a certain element).
3. Case studies
2.8.4. Categorizing The many measurement types and acquisition modes of
Unlike the filters that are used to exclude particles/data the TIMA system were tested at the Institute of Geology
from charts, tables and visualization the categorizers of the Czech Academy of Sciences in numerous scientific
are employed to subdivide the data into user-defined projects in areas spanning from ore geology (Haluzová
categories (classes). Each category is defined by a set of et al. 2015), through paleontology and paleoecology
rules, based on properties of the data objects (e.g. par- (Slavík et al. 2016), petrology (Svojtka et al. 2016; Žák
ticle size, liberation, phase contents, elemental content et al. 2016; Ackerman et al. 2017; Breiter et al. 2017,
etc.). Combinations of rules of different kind are also 2018), archaeology (Neumannová et al. 2016), ecol-
possible. Categories are now used only in specific report ogy/contaminated soil analysis (Harvey et al. 2017) to
types. A simple example would be to create individual dust analysis (Hrstka et al. 2017b). The majority of the
sub-groups of particles based on their textural properties applications have been linked to the fast collection of
like liberation or association. (e.g., Area % (Min- statistically robust data on mineral composition and tex-
eral)>=80 <=100, Area % (Mineral)>=60 tural characteristics, together with the ability to quickly
<=80, Area %(Mineral)>=20 <=60, Area search for specific minerals or phases within geological
% (Mineral)>0 <=20). For particulate samples this samples. To illustrate the novel approach of Dot map
would create classes of nearly free (liberated) mineral of (TDM) analysis and Bright phase search (TBPS) in thin
interest, relatively liberated mineral of interest, mixed sections (specific mineral search), selected case studies
particles composed of mineral of interest and other on applications of the TIMA AM technology are briefly
phases and particles with low proportion of the mineral of outlined in this section.
interest. While liberation or association information are
mostly used in mineral processing to model the theoreti- 3.1. Case study 1 – fast automated search
cal behavior of particles in a physical separation unit like for specific minerals in thin and polished
cyclone or flotation cell (Sandmann and Gutzmer 2013; sections
Jordens et al. 2016) the same concept can be easily ex-
tended to classifying the rock fragments or thin sections The modal analysis in combination with Bright phase
according to their petrological classification (“lithotyp- search (TBPS) was used to provide better constraints
ing”) (Haberlah et al. 2010; Knappett et al. 2011; Higgs on the nature, conditions of formation and evolution of
et al. 2015). Automated classification according to the sulphide mineralization at Rožany deposit, northernmost
QAPF classification scheme for igneous rocks (Le Maitre Bohemian Massif (Haluzová et al. 2015). Studied sam-
2002) can be an excellent example. As we can use the ples were collected in two abandoned quarries (the first

57
Tomáš Hrstka, Paul Gottlieb, Roman Skála, Karel Breiter, David Motl

quarry: N 51°2.09'; E 14°27.08', the second: N 51°2.05'; Also the spatial distribution of magnetite in the samples
E 14°27.87') in 2014–2015. In total, six samples of mas- was used to shed some light on the magnetostratigraphic
sive ores, two of disseminated ores and one of the barren properties of the rocks.
dolerite were selected from a large sample collection. The TIMA has found over a thousand monazite grains,
These were used to investigate the mineralogy and char- 1–80 µm across, some displayed a complex inner structure
acteristics of the Ni–Cu–(PGE) mineralization. formed by visible relict nuclei crystals or crystoclasts rich
The distribution of the individual platinum-group in Th, surrounded by altered overgrowths with a spongy
minerals (PGM) was established through the detailed or crystalline appearance considerably depleted in Th.
“TBPS” analysis. For the eight studied samples, over The outermost rims were again enriched in Th. From the
29,000 potential target grains (0.5–200 µm in size) were detailed analysis of individual monazite grains, the identi-
identified based on the BSE brightness. By matching fied textures and comparison to similar material from other
the EDS signature to the classification scheme, 63 PGM locations worldwide, understanding on the paleoclimate and
grains were located in less than 10 h. Their identifica- its evolution was improved (Slavík et al. 2016 and refer-
tion through the automated system was crosschecked by ences therein). Based on this published example, automated
a detailed manual SEM/EMPA investigation of selected mineralogy data together with systematic specific mineral
individual grains located through the TIMA analysis. search and related deportment of elements of interest in
The correlative TIMA workflow allowed exporting the thin sections and rock samples offers wide and promising
co-ordinates of target grains into the EPMA system. In applications in identification of source areas, paleoclimate
addition to the previous studies, minerals moncheite reconstructions and provenience studies in general.
[(Pt,Pd)(TeBi) 2] and michenerite [(Pd,Pt)BiTe] were
revealed and a possible unnamed Pt–As–Te phase was
also noted. During the TIMA investigation the PGM were 3.3. Case study 3 – Cínovec/Zinnwald Pluton,
found not only directly in the base-metal sulphides or Krušné Hory Mts./Erzgebirge
at sulphide–silicate grain boundaries, but also in matrix The 1597 m deep borehole CS-1 located in the center of
silicates, amphiboles and/or chlorites. This indicates that the Cínovec Pluton represents an ideal object for evalua-
at least part of the PGM was related to the late-stage tion of vertical structure of rare-metal bearing magmatic
hydrothermal processes. systems (Štemprok and Šulcek 1969; Štemprok 2016;
The distribution of the individual PGE was shown to Breiter et al. 2017).
correlate with the likely incorporation of Iridium group A combination of textural and chemical methods was
of PGM (I-PGM) in Ni-bearing sulphides in the form applied to the whole-rock and mineral samples to define
of solid solution. The results are in agreement with the chemical and mineral composition of all granite facies and
findings of Sandmann and Gutzmer (2015) indicating to assess the relative role of magmatic and metasomatic
that high number of polished sections needs to be studied processes during differentiation of the pluton and forma-
to provide statistically valid data on PGM distribution. tion of the Cínovec/Zinnwald Li–Sn greisen deposit.
The automated approach enabled fast localization and One of the tasks was the estimation of temperature of
identification of PGM among the BSE-bright phases crystallization of the main granite facies using feldspar
within the samples. It also helped to assess the economic geothermometer (Whitney and Stormer 1977).
potential of the deposit and to constrain the ore genesis The Cínovec Pluton is composed of two principal
in comparison with other Ni–Cu–(PGE) mineralizations units: a suite of albite–zinnwaldite granites forms upper
(Haluzová et al. 2015). part to the depth of 735 m, whereas the suite of biotite
granites follows in interval 735–1597 m. Both suites
3.2. Case study 2 – palaeoecological locally comprise fine-grained distinctly porphyritic fa-
characterization of the Lochkovian– cies (traditionally termed as microgranites) with perthite
Pragian boundary phenocrysts 5–10 mm in size.
To calculate the primary perthite composition and es-
A highly automated approach was used in the paleoeco- timate the temperature of its crystallization, typical sam-
logical characterization of the sedimentary record across ples of both microgranites (zinnwalditic from the depth
the Lochkovian–Pragian boundary in the Spanish Central of 413 m and biotitic from the depth of 880 m) were
Pyrenees to explain the origin of the anomalously high selected. The surface of thin section was divided into a
(17 to 26 ppm) Th contents in this stratigraphic interval square grid (c. 300 squares per a standard section) and the
as well as to better understand the potential source areas abundances of Ab and Kfs in each square were measured
and provenance of impurities in the carbonates (Slavík et using TIMA. The results of the squares composed only
al. 2016). In this case the use of AM allowed us to trace of perthite were highlighted, and carefully inspected to
the source of Th and identify monazite as its main carrier. verify if all of the albite really represented perthite ad-

58
 Automated mineralogy and petrology – applications of TESCAN Integrated Mineral Analyzer (TIMA)

5 Kfs
Ab
Qz
4 Tur
Ms
Znw
Ap
3 Tpz
Rest
2

a 5 mm b 5 mm
1 c
0 10 20 30 40 50 60 70 80 90 100 vol.%
Fig. 7 Detailed investigation of a fine-grained layered rock (Megiliggar, Cornwall): a – Macro-photo; b – Mineral map by TIMA; c – Modal com-
position of individual layers. Mineral abbreviations after Whitney & Evans (2010). Crystallization of the rock proceeded upwards.

mixtures. After the exclusion of all squares containing body in some neighboring granite pluton, is the topic
also the likely primary albite, the mean composition of of a long-term debate (London 2008). Despite intense
perthite within the section was calculated. The primary search, localities enabling to study the direct transition
composition of the alkali feldspars was obtained on the from granite pluton to rare-element pegmatite dykes are
basis of modal abundances of the K- and Na-phases and scarce. One of the best examples is the Megiliggar Rock
their chemical compositions measured by the EMPA. The at the SE contact of the Tregonning Granite in Cornwall,
calculated Ab component in primary feldspar decreases SW England (Stone 1969). Spectacular outcrops in the
from c. 40 % in the biotite microgranite (BtGm) at the coastal cliff show the transition from the granite pluton
depth of 860 m to c. 21 % in the zinnwaldite microgranite through dyke leucogranites to aplites and pegmatites.
(ZiGm) at the depth of 413 m. Considering the decrease The TIMA was used for a detailed study of the modal
in pressure during magma ascent in the Eastern Erzge- and chemical compositions of layered parts of the dykes,
birge (6 → 1 kbar; Müller et al. 2005) and the primary where the thickness of individual layers varies between
composition of associated plagioclase (An15 in the BtGm 1 and 5 mm (Breiter et al. 2018).
and An05 in the ZiGm), the temperature of the perthite Mineral maps across the zoning were constructed and
equilibration decreases upwards from approximately mineral compositions in texturally distinct layers com-
800 °C to 550 °C. In this case the automated collection of puted using the TIMA technology (Fig. 7). Prior map-
modal and textural data allowed better understanding of ping, typical EDS spectra of all analysed minerals were
the granite evolution as detailed in Breiter et al. (2017). collected directly from the studied thin sections which
substantially improved classification of different varieties
of micas and other less common minerals.
3.4. Case study 4 – estimation of mineral
Modal composition from part of one of the measured
and chemical composition of layered
profiles is shown in Tab. 2 and Fig. 7. Modal composition
aplite–pegmatite dykes: “line rocks”
of the sequence evolved systematically by the decrease in
of Megiliggar, Cornwall
K-feldspar (28→6.5 vol. %), quartz (33→10 vol. %), and
Genetic relation between granitic pegmatites and gran- muscovite (12→2 vol. %), along with a strong increase
ites, i.e. a search for the source of particular pegmatite in albite (16→72 vol. %). Contents of Qtz, Kfs and Ms

59
Tomáš Hrstka, Paul Gottlieb, Roman Skála, Karel Breiter, David Motl

Tab. 2 Modal and computed chemical compositions of the individual aplite layers from Megiliggar systematically decrease, while
(Cornwall) (vol. % resp. wt. %) the content of Ab increases
Layer # 1 2 3 4 5 6 7 upwards in the profile, i.e. in
Orthoclase 27.86 18.18 22.60 19.81 11.90 8.39 6.52 the direction of crystallization.
Albite 16.39 27.94 33.79 34.67 49.02 59.77 72.43 Within the Qtz–Ab–Kfs triangle
Quartz 32.96 35.74 31.16 33.53 25.24 18.97 9.90 (Fig. 8), a shift of the rock com-
Schorl 8.14 5.68 7.13 6.36 8.75 8.57 4.33 position from Qtz + Kfs-domi-
Muscovite 11.77 9.08 2.57 2.10 1.65 1.16 2.00 nated to Ab-dominated, i.e. in
Zinnwaldite 0.11 0.09 0.14 0.15 0.33 0.21 0.30 direction of common evolution
Apatite 0.68 0.81 0.55 0.54 0.60 0.29 1.12 of leucogranitic melts, is clear.
Topaz 0.40 0.51 0.00 0.01 0.01 0.00 0.01 Approximate chemical com-
Fluorite 0.06 0.09 0.09 0.10 0.11 0.12 0.16
positions of individual layers
Rutile 0.00 0.00 0.00 0.02 0.00 0.00 0.00
were computed from modal
Columbite 0.00 0.00 0.00 0.00 0.00 0.00 0.00
compositions and chemistries of
Zircon 0.00 0.01 0.01 0.01 0.01 0.02 0.00
Sum 98.37 98.13 98.04 97.29 97.62 97.50 96.79
rock-forming minerals analyzed
SiO2 71.06 73.34 72.99 73.65 70.92 69.41 66.64
by the microprobe (Li by LA-
TiO2 0.15 0.11 0.06 0.06 0.06 0.05 0.04 ICP MS) (Tab. 2). The chemi-
Al2O3 15.15 13.92 13.81 13.07 15.10 16.27 17.22 cal composition of the whole
FeOt 1.36 0.97 1.03 0.92 1.25 1.20 0.66 layered sequence should be cal-
MgO 0.05 0.04 0.04 0.04 0.05 0.05 0.03 culated as a mean of the com-
MnO 0.04 0.04 0.03 0.03 0.03 0.02 0.04 positions of individual layers,
CaO 0.42 0.52 0.40 0.40 0.46 0.32 0.79 taken into account their relative
Li2O 0.00 0.00 0.00 0.01 0.01 0.01 0.01 thickness. This method enabled,
Na2O 2.17 3.45 4.19 4.27 5.99 7.24 8.65 among others, to compute con-
K 2O 6.06 4.12 4.17 3.64 2.28 1.63 1.43 tents of chemical elements dif-
Rb2O 0.08 0.06 0.06 0.05 0.03 0.02 0.02 ficult to analyze, such as boron.
P 2O 5 0.31 0.37 0.26 0.26 0.28 0.16 0.51 The automated approach with
B 2O 3 0.81 0.56 0.71 0.63 0.87 0.85 0.43 fast modal analysis and good
F 0.26 0.26 0.09 0.09 0.11 0.09 0.13
textural resolution significantly
Sum 96.85 96.95 97.04 96.39 96.46 96.38 96.03
shortened the time needed for
mapping of large areas of the
Qtz petrological samples.

0
100 4. Discussion and future
10    developments
90
20 The goal of the ongoing TIMA-
80 X development is to provide
30 a comprehensive automated
70
40 mineralogy tool with increasing
60 level of automation, through
50 the integration of artificial in-
2 50 telligence and expert-guided
60 1 machine learning. The current
40 system is a highly specialized
70 5 4
3 automated SEM providing ex-
30 cellent tool for geosciences and
80 6
20 material science in general. For
90
7 10 Fig. 8 Composition of individual layers
100 of lined rocks (Megiliggar, Cornwall)
0 in the Ab– Qtz–Kfs triangle. Note the
systematic shift to the Ab-apex from
Ab 0 10 20 30 40 50 60 70 80 90 100 Kfs the point 1 to 7.

60
 Automated mineralogy and petrology – applications of TESCAN Integrated Mineral Analyzer (TIMA)

further improvements of the AM technology there are still AV21/4 project and with help of Institute of Geology of
number of topics which can be addressed. the CAS, v. v. i. research plan RVO 67985831. Finally,
From the instrumentation perspective: further speeding we would like to thank SGS Minerals for their support
up of data acquisition, and better integration of additional of this study.
detectors providing extended input of information (Ra-
man, WDS, cathodoluminescence, and micro-XRF) might Electronic supplementary material. Supplementary data
be useful. on mean atomic numbers of the selected minerals for
From the software perspective: parallelization in data this paper are available online at the Journal website
processing and integration of big data concepts could (http://dx.doi.org/10.3190/ jgeosci.250).
provide significant advantage while working on projects
with hundreds or thousands of samples. Also the correla-
tive microscopy approach helping to integrate data from References
other techniques and above-mentioned potential new
detectors will be highly beneficial. In the short term, the Ackerman L, Magna T, Rapprich V, Upadhyay D, Krátký
integration of more sophisticated textural parameters is O, Čejková B, Erban V, Kochergina YV, Hrstka
under development. T (2017) Contrasting petrogenesis of spatially related
Automated classification file construction through the carbonatites from Samalpatti and Sevattur, Tamil Nadu,
means of artificial intelligence, neural networks, principal India. Lithos 284: 257–275
component analysis (PCA), machine learning, similar- Altree-Williams A, Pring A, Ngothai Y, Brugger J
ity search and hierarchical clustering is also a matter (2015) Textural and compositional complexities result-
of interest (Hrstka et al. 2017a). The implementation of ing from coupled dissolution–reprecipitation reactions
standards for cross-correlation between different instru- in geomaterials. Earth Sci Rev 150: 628–651
ments would be also highly desirable in the future. This Andersen JCØ, Rollinson GK, Snook B, Herrington R,
seems particularly important as more AM systems are Fairhurst RJ (2009) Use of QEMSCAN® for the char-
being developed and there has been very little research acterization of Ni-rich and Ni-poor goethite in laterite
comparing their analytical capabilities so far. ores. Miner Eng 22: 1119–1129
Berrezueta E, Ordóñez-Casado B, Bonilla W, Banda
R, Castroviejo R, Carrión P, Puglla S (2016) Ore
5. Conclusions petrography using optical image analysis: application
to Zaruma-Portovelo Deposit (Ecuador). Geosciences
Combined use of all available signals (mainly BSE and 6: DOI 10.3390/geosciences6020030
EDS) in conjunction with advanced image and pattern- Breiter K, Ďurišová J, Hrstka T, Korbelová Z, Vaňková
recognition analysis involving artificial intelligence are M, Vašinová Galiová M, Kanický V, Rambousek P,
required for any modern automated SEM-based mineral Knésl I, Dobeš P, Dosbaba M (2017) Assessment of
analysis system. Such systems can be used not only in their magmatic vs. metasomatic processes in rare-metal gran-
classic domain of mineral processing, plant monitoring, ites: a case study of the Cínovec/Zinnwald Sn–W–Li
optimization and design but can provide valuable informa- deposit, Central Europe. Lithos 292/293: 198–217
tion in a broad variety of Earth and material sciences. We Breiter K, Ďurišová J, Hrstka T, Korbelová Z, Vaňková
have demonstrated the high potential of TIMA in extending M, Vašinová Galiová M, Müller A, Simons B, Shail
the everyday tools for geoscience research by a number of RK, Williamson BJ, Davies JA (2018) The transition
case studies. We have also shown that the new generation from granite to banded aplite–pegmatite sheet complexes:
of automated mineralogy instruments builds on the origi- an example from Megiliggar Rocks, Tregonning topaz
nal concepts introduced in the early 1970’s but provides granite, Cornwall. Lithos 302/303: 370–388
additional benefits due to the improvements in analysis Donskoi E, Suthers SP, Fradd SB, Young JM, Campbell
modes, full integration of quantitative EDS and availability JJ, Raynlyn TD, Clout JMF (2007) Utilization of opti-
of computation power and complex programing. cal image analysis and automatic texture classification for
iron ore particle characterisation. Miner Eng 20: 461–471
Acknowledgements. We are indebted to Rolando Lastra, Fandrich R, Gu Y, Burrows D, Moeller K (2007) Mod-
Jens Gutzmer, Rogerio Kwitko Ribeiro and the editors ern SEM-based mineral liberation analysis. Int J Miner
for the review. Their comments and suggestions on the Process 84: 310–320
manuscript greatly improved the final text. We would Glagolev AA (1934) Quantitative analysis with the mi-
also like to thank TESCAN for the support of the TIMA croscope by the point method. Eng Min J 135: 399–400
application laboratory at Institute of Geology of the CAS, Gottlieb P (2008) The revolutionary impact of automated
v. v. i. This study was financed also through Strategie mineralogy on mining and mineral processing. In: Wang

61
Tomáš Hrstka, Paul Gottlieb, Roman Skála, Karel Breiter, David Motl

DZ, Sun CY, Wang FL, Zhang LC, Hang L (eds) Johnson C, Pownceby MI, Wilson NC (2015) The appli-
24th International Mineral Processing Congress. Science cation of automated electron beam mapping techniques
Press, Beijing, pp 165–174 to the characterisation of low grade, fine-grained min-
Gottlieb P, Wilkie G, Sutherland D, Ho-Tun E, Suthers eralisation; potential problems and recommendations.
S, Perera K, Jenkins B, Spencer S, Butcher A, Rayner Miner Eng 79: 68–83
J (2000) Using quantitative electron microscopy for pro- Jordens A, Marion C, Grammatikopoulos T, Waters
cess mineralogy applications. JOM 52: 24–25 KE (2016) Understanding the effect of mineralogy on
Haberlah D, Williams MAJ, Halverson G, McTainsh muscovite flotation using QEMSCAN. Int J Miner Pro-
GH, Hill SM, Hrstka T, Jaime P, Butcher AR, Glasby cess 155: 6–12
P (2010) Loess and floods: high-resolution multi-proxy Knappett C, Pirrie D, Power MR, Nikolakopoulou
data of Last Glacial Maximum (LGM) slackwater deposi- I, Hilditch J, Rollinson GK (2011) Mineralogical
tion in the Flinders Ranges, semi-arid South Australia. analysis and provenancing of ancient ceramics using
Quat Sci Rev 29:2 673–2693 automated SEM-EDS analysis (QEMSCAN®): a pilot
Haluzová E, Ackerman L, Pašava J, Jonášová Š, Svojtka study on LB I pottery from Akrotiri, Thera. J Archaeolog
M, Hrstka T, Veselovský F (2015) Geochronology Sci 38: 219–232
and characteristics of Ni–Cu–(PGE) mineralization at Kwitko-Ribeiro R (2012) New sample preparation de-
Rožany, Lusatian Granitoid Complex, Czech Republic. velopments to minimize mineral segregation in process
J Geosci 60: 219–236 mineralogy. In: Broekmans MATM (ed) Proceedings of
Harding DP (2002) Mineral identification using a scanning the 10th International Congress of Applied Mineralogy.
electron microscope. Miner Metall Process 19: 215–219 Springer, Berlin, Heidelberg, pp 411–417
Harvey PJ, Rouillon M, Dong C, Ettler V, Handley Lane GR, Martin C, Pirard E (2008) Techniques and ap-
HK, Taylor MP, Tyson E, Tennant P, Telfer V, Trinh plications for predictive metallurgy and ore characteriza-
R (2017) Geochemical sources, forms and phases of soil tion using optical image analysis. Miner Eng 21: 568–577
contamination in an industrial city. Sci Total Environ Larrea ML, Castro SM, Bjerg EA (2014) A software
584–585: 505–514 solution for point counting. Petrographic thin section
Higgs KE, Haese RR, Golding SD, Schacht U, Watson analysis as a case study. Arabian J Geosci 7: 2981–2989
MN (2015) The Pretty Hill Formation as a natural ana- Lastra R (2007) Seven practical application cases of libera-
logue for CO2 storage: an investigation of mineralogical tion analysis. Int J Miner Process 84: 337–347
and isotopic changes associated with sandstones exposed Lastra R, Paktunc D (2016) An estimation of the variabil-
to low, intermediate and high CO2 concentrations over ity in automated quantitative mineralogy measurements
geological time. Chem Geol 399: 36–64 through inter-laboratory testing. Miner Eng 95: 138–145
Hoal KO, Stammer JG, Appleby SK, Botha J, Ross JK, Le Bas MJ, Streckeisen AL (1991) The IUGS systemat-
Botha PW (2009) Research in quantitative mineral- ics of igneous rocks. J Geol Soc, London 148: 825–833
ogy: examples from diverse applications. Miner Eng Le Maitre RW (2002) Igneous Rocks: A Classification
22: 402–408 and Glossary of Terms. Cambridge University Press,
Hrstka T (2008) Preliminary results on the reproducibility Cambridge, pp 1–236
of sample preparation and QEMSCAN® measurements Lloyd GE (1987) Atomic number and crystallographic
for heavy mineral sands samples. Ninth International contrast images with the SEM: a review of backscattered
Congress for Applied Mineralogy, proceedings: ICAM electron techniques. Mineral Mag 51: 3–19
2008. Australasian Institute of Mining and Metallurgy, London D (2008) Pegmatites. Canad Mineral, Spec Publ
Carlton (Vic), pp 107–111 10: pp 1–347
Hrstka T (2012) Paleofluid Chemistry of Orogenic Gold Lotter NO, Kormos LJ, Oliveira J, Fragomeni D, White-
Deposits: Novel Analytical Methods and Case Studies man E (2011) Modern process mineralogy: two case
from the Bohemian Massif. Unpublished PhD thesis, studies. Miner Eng 24: 638–650
Faculty of Science, Charles University in Prague, pp Mineralogical database (2017) Mineralogical database.
1–151 Accessed on May 10, 2017 at http://www.webmineral.com
Hrstka T, Gottlieb P, Moravec J, Lokoc J (2017a) The Motl D, Filip V (2013) Method and apparatus for material
future of SEM-based automated mineralogy and artifi- analysis by a focused electron beam using character-
cial intelligence in applied mineralogy. Sci Res Abstr 6: istic X-rays and back-scattered electrons. (Patent no.
Digilabs, pp 42 WO2017050303 A1)
Hrstka T, Motl D, Gottlieb P, Hladil J (2017b) Using Müller A, Breiter K, Seltmann R, Pécskay Z (2005)
an automated approach in building a dust particle atlas Quartz and feldspar zoning in the eastern Erzgebirge
for research and environmental monitoring. Sci Res Abstr volcano–plutonic complex (Germany, Czech Republic):
6: Digilabs pp 43 evidence of multiple magma mixing. Lithos 80: 201–227

62
 Automated mineralogy and petrology – applications of TESCAN Integrated Mineral Analyzer (TIMA)

Neumannová K, Thér R, Květina P, Hrstka T (2016) Slavík L, Valenzuela-Ríos JI, Hladil J, Chadimová
Perception of technological variability on the example L, Liao JC, Hušková A, Calvo H, Hrstka T (2016)
of Neolithic settlement site in Bylany. Praehistorica 33: Warming or cooling in the Pragian? Sedimentary record
291–306 (in Czech) and petrophysical logs across the Lochkovian–Pragian
Nie J, Peng W (2014) Automated SEM–EDS heavy mineral boundary in the Spanish Central Pyrenees. Palaeogeogr
analysis reveals no provenance shift between glacial loess Palaeoclimatol Palaeoecol 449: 300–320
and interglacial paleosol on the Chinese Loess Plateau. Smith DGW, Nickel EH (2008) Codification of unnamed
Aeolian Res 13: 71–75 minerals. J Petrol 49: 581–583
Pérez-Barnuevo L, Pirard E, Castroviejo R (2013) Smythe DM, Lombard A, Coetzee LL (2013) Rare Earth
Automated characterisation of intergrowth textures in Element deportment studies utilising QEMSCAN tech-
mineral particles. A case study. Miner Eng 52: 136–142 nology. Miner Eng 52: 52–61
Pirard E (2004) Multispectral imaging of ore minerals in Stone M (1969) Nature and origin of banding in the granitic
optical microscopy. Mineral Mag 68: 323–333 sheets Tremearne, Porthleven, Cornwall. Geol Mag 106:
Pirrie D, Rollinson GK (2011) Unlocking the applications 142–158
of automated mineral analysis. Geol Today 27: 226–235 Svojtka M, Ackerman L, Medaris LG, Hegner E, Valley
Poliakov A, Donskoi E (2014) Automated relief-based JW, Hirajima T, Jelínek E, Hrstka T (2016) Petrologi-
discrimination of non-opaque minerals in optical image cal, geochemical and Sr–Nd–O isotopic constraints on
analysis. Miner Eng 55: 111–124 the origin of garnet and spinel pyroxenites from the
Pooler R, Dold B (2017) Optimization and quality control Moldanubian Zone of the Bohemian Massif. J Petrol
of automated quantitative mineralogy analysis for acid 57: 897–920
rock drainage prediction. Minerals 7: DOI 10.3390/ Štemprok M (2016) Drill hole CS-1 penetrating the Cí­
min7010012 novec/Zinnwald granite cupola (Czech Republic): an
Rollinson GK, Andersen JCØ, Stickland RJ, Boni M, A-type granite with important hydrothermal mineraliza-
Fairhurst R (2011) Characterisation of non-sulphide tion. J Geosci 61: 395–423
zinc deposits using QEMSCAN®. Miner Eng 24: 778–787 Štemprok M, Šulcek Z (1969) Geochemical profile
Sandmann D, Gutzmer J (2013) Use of mineral liberation through an ore-bearing lithium granite. Econ Geol 64:
analysis (MLA) in the characterization of lithium-bearing 392–404
micas. J Miner Mater Charact Eng 1: 285–292 Ward I, Merigot K, McInnes BIA (2018) Application of
Sandmann D, Gutzmer J (2015) Nature and distribution Quantitative Mineralogical Analysis in archaeological
of PGE mineralization in gabbroic rocks of the Lusatian micromorphology: a case study from Barrow Is., Western
Block, Saxony, Germany. Z Geol Wiss 166: 35–53 Australia. J Archaeol Method Theory 25: 45–68
Sánchez E, Deluigi MT, Castellano G (2012) Mean Whitney JA, Stormer JC (1977) The distribution of NaAl-
atomic number quantitative assessment in backscattered Si3O8 between coexisting microcline and plagioclase and
electron imaging. Microsc Microanal 18: 1355–1361 its effect on geothermometric calculations. Amer Miner
Santoro L, Boni M, Rollinson GK, Mondillo N, Balas- 62: 687–691
sone G, Clegg AM (2014) Mineralogical characteriza- Whitney DL, Evans BW (2010) Abbreviations for names
tion of the Hakkari nonsulfide Zn(Pb) deposit (Turkey): of rock-forming minerals. Amer Miner 95: 185–187
the benefits of QEMSCAN®. Miner Eng 69: 29–39 Žák K, Skála R, Řanda Z, Mizera J, Heissig K, Acker-
Santoro L, Rollinson GK, Boni M, Mondillo N (2015) man L, Ďurišová J, Jonášová Š, Kameník J, Magna
Automated Scanning Electron Microscopy QEMSCAN®- T (2016) Chemistry of Tertiary sediments in the sur-
based mineral identification and quantification of the roundings of the Ries impact structure and moldavite
Jabali Zn–Pb–Ag nonsulphide deposit (Yemen). Econ formation revisited. Geochim Cosmochim Acta 179:
Geol 110: 1083–1099 287–311

63