Sensor-based ore sorting methodology

investigation applied to gold ores

DUMONT, J.A.1, LEMOS, M. G. 2 , ROBBEN, C.
1. Mining and Technical, Anglogold Ashanti, Brazil
2. Project and Industrial Process, Anglogold Ashanti, Brazil
3. Mining, TOMRA Sorting, Germany

ABSTRACT

Exploring gold ores has become more challenging every year. Mines have increasingly delivered
lower grades and complex mineralogy, which forces mills to process more material to reach the
production plan.
Research, testing, and implementation of innovative technologies can prevent high capital
investments in metallurgical plants, while maintaining or even increasing production targets even at
lower grades. The ore sorting technology provides a pre-concentration of the ore, in the early stages
of the production process. The pre-concentration is based on X-ray-transmission (XRT), laser, optical,
infrared and/or electromagnetic sensors analysis to identify, and air ejections to separate ore from
sub-economic waste.
The aim is to increase mine recovery by reducing cut-off grade and process the extra material without
compromising the current mine production, opening new capacity in the mill by rejecting waste
material, without significantly amount of capital expended.
The understanding of the relationship between different texture types, grain size and the gold source
associations such as quartz or sulphide, and their responses to different types of sensors are the key
to project success. The methodology consists to identify the main textures, clustering by the rock
description, understanding the response behaviour of each typologies to the sensors, chemical and
mineralogical analysis by textures to identify the gold sources, the machine configuration with the
information obtained in the last step and pilot test. In order to achieve the objectives, research has
been developed to define the types of calibration and sensors for the different types of gold ore
produced by Anglogold Ashanti.
The methodology showed potential to achieve excellent gold recovery (higher than 95%) in/with? a
mass pull of 40%, using one or two steps of pre-concentration. This information confirms the
importance of a geological approach when investigating ore sorting technology for complex gold
ores. In addition, the information acquired supports the stability of other processes such as
metallurgy and mine geology.

1
INTRODUCTION

The sorting technology provides a pre-concentration of the ore, in the early stages of the production
process. The major aim is to increase mine recovery by reducing cut-off grade and make feasible new
mine areas (low grade or high dilution zones) without compromising the currently mine plan, adding
new production to the plan. Generally, ore sorting equipment is operated with run-of-mine (ROM)
material as feed, after two-step crushing which corresponds to a relatively low-cost operation. In
addition to the production increase, the implementation of ore sorting can be a cheaper alternative if
compared with a plant expansion, and provides a lower operational cost and lower tailings
management cost.
Due to the potential of this technology, AngloGold Ashanti (AGA) conducted an extensive testing
program with sensor-based ore sorting technology using distinct ROM grades and types from the
different mines.
The Bench-Scale Tests and Performance Tests were carried out in TOMRA’s Test Centre in Germany
and showed a potential for application in gold ores. From these results, it was decided to set up a
pilot plant, using a XRT sorting machine, in one of the AngloGold Ashanti mines in Brazil, where
samples from different mines were sent to be tested.

Ore Sorting Technology - Concepts

The basic principle of ore sorting equipment is to identify the metal of interest in each rock particle
that passes through the machine, and then separate them from the particles that do not contain the
metal of value (or such small quantities, that it is economically not viable to further process). The
crushed rocks pass through a certain sensor and the signal from the sensor is sent to a processor,
which decides whether the particle contains the metal of value and is present in economically
recoverable quantities. This decision is transmitted to the physical separation mechanism, which
takes the action to form two fractions of particles: one that contains the metal of value and another
that does not (or in too low concentrations to be processed economically feasible).

2
 Figure 1 Ore sorting mechanism, belt-type XRT sorting machine (TOMRA).

The main properties detected for separation between product and waste fractions are colour, atomic
number, electromagnetic susceptibility, laser diffraction and others. Nowadays, sensor-based ore
sorters are widely used in industrial minerals operations, diamond mines and precious and non-
precious metal mines.
The automated classification applies optical sensors (Color), laser detection systems (Laser), near
infrared spectroscopy (NIR) or X-ray transmission (XRT), which may or may not be coupled with
sensors of electrical conductivity and magnetic susceptibility (EM), to control the mechanical
separation of ore in two or more fractions. Table 1 shows the main textures present in the lithotypes
found in AngloGold Ashanti mines and possible sensors that can be applied for pre-concentration.

Table 1 Sensors and their application for AGA textures

Texture (Gold Source) Sensors

Quartz/Smoke Quartz with sulphides Laser, XRT, Color

Disseminated and fine sulphides XRT

associated with schist and gneisses

3
 Disseminated and fine sulphides Color, NIR, XRT, EM (mainly pyrrhotite)
associated with chlorite schist and mafic
rocks

Banded Iron Formation XRT, EM

Altered schists with residual sulphides Color, XRT, NIR

Massive sulphides XRT

In the XRT machine installed for testing in the AngloGold Ashanti pilot plant, two technologies are
applied in the detection stage for classification: Dual Energy X-ray transmission (DE-XRT) and single
X-ray signal (for inclusion analysis). For both technologies, the basic operation consists of emission
of an x-ray beam on the sample, transmission through the matter, attenuation thereof and detection
of the residual radiation transmitted. The radiation in each pixel of the image converts into an
electrical signal, which is then set in a grayscale image. The dots or dark portions in the image
represent compositions of high absorption and consequently high density. For DE-XRT, two XRT
channels are used to analyse the thickness and density of the material. To detect small inclusions just
one energy channel is used. The Figure 2 shows some ores textures and the respond of each
technologies using the XRT machine.

Figure 2 XRT electronic images on grayscale (1) and processed images using DE-XRT (2) and inclusion analysis
(3) for different particle textures

METHODOLOGY

The first step of the tests is the size classification stage, where the size ranges are defined to be feed
in the machine. It is a decision based mostly in the liberation, sensor capability and particle size
distribution. The typical size ranges selected to be sorted were 10-20mm and 20-45mm. Both sizes are
investigated separately in the steps below. The fines (below 10mm) cannot be sorted and has its
destination based on its grade.

4
To investigate the Ore Sorting capability, it is necessary to follow steps to understand the Au carrying
typologies and ore types (lithotype/ texture/ phases) for each deposit / mine / ore body for each grain
size (Figure 4). This methodology consists out of six steps of investigation: Sampling (1) Hand Sorting
(2), Bench-Scale (Rock- by-Rock) Test (3), Lithology Test (4), Performance Test (5) and Pilot Test (6),
according to flow chart below.

Figure 4 Workflow of the ore sorting tests

5
Sampling
Sampling is of high importance for the methodology. As all following steps are labour and cost
intensive, it is important to conduct the work on carefully selected samples that represent the
variability of the ore body in question. For the first three test phases Hand Sorting, Bench-Scale Test
and Lithotype Test, the samples are composited to represent each lithotype with an equal amount of
sample (not representative for the mass/ frequency distribution within the deposit). The sample set
must cover all variations that may be a portion of future production. Calibration and validation sets
must be taken individually to prevent overfitting.
Samples for Performance Tests and Piloting activities need to represent the composition of future
mine production and must therefore be representative for the mass-frequency distribution of the
lithotypes.

Hand Sorting
The starting point in the investigation is to understand the gold carrier phases for each deposit / mine
for each grain size. It is necessary to classify and quantify the different typologies and ore types
(lithotype/ textures/ phases) seeking for a previous understanding and elaboration of the working
algorithm for the analysis stage. The classification step by type is defined by hand sorting and is
performed with the support of a geologist and geology technician. The sample is divided in several
lithotype / ore type classes which will be tested individually in the steps below.

Bench-Scale (Rock-by-Rock) Test / Calibration

The Bench-Scale Test aims to identify which textures are sources of gold and what their responses
are to the sensors studied. The rock fragments should be individually tested according to their
textural class in sensor-based ore sorting and gold and sulphur analyses, and another chemical
element(if necessary) will be performed. From the information acquired, programs and new
configurations will be developed and thus the sensor-based ore sorter initially calibrated before being
validated in the next steps.

Lithotype Test/ Validation

The lithotype tests are the first stages to validate the configurations and the programs made in the
previous stage with an independent sample set. The machine is feed with small samples (10 to 20 kg
each) with material from a specific lithotype. The signals from each particle are recorded and used as
dataset to further program optimization. The efficiency of the program for this lithotype are validated
based in the ejection mass and chemical analysis (gold and sulphur) of the products again on new
sample sets to avoid cross validation and overfitting. The erroneous classifications will be analysed
to understand the variability of the classes defined before, to measure the classification efficiency,
adjustments of the sensitivity and ejection parameters, defined logic and errors in the program.

6
Performance Test
The Performance Tests are the stages to validate the programs’ parameters setup. In addition to the
so-called Detection Efficiency, other (mechanical) sub-process efficiencies now contribute to the Total
Process Efficiency of the sensor-based ore sorting stage (Robben, 2013). Fine-tune is made to adjust
the sensitivity of the separation and define the products zones defined. It is an import step to identify
the optimized operational setup and it is the first contact of the machine with the full sample without
any previous separation. This step provides the mass balance and grades to be achieved in the
process. The erroneous classifications in the products will be analysed to identify any potential
adjustment.
Pilot Test
In the pilot tests, the programs selected are already adjusted, optimized and defined. The main
objective is to validate the repeatability and variability of the previous steps on larger samples
representing the heterogeneity and variability of future production. It can be run at 40t/h in a steady
operation and sampled with an automated system in a short interval between increment extractions.
Product belts have weightometers installed. Operational parameters and process control variables
can also been defined, as optimized throughput, material handling, resource input consumptions,
etc.

RESULTS AND DISCUSSION

The sensor-based ore sorting investigation methodology applied, as described in the previous
section, were carried out with several ROM samples from AngloGold Ashanti Brazil mines. In this
article the results of test work for one selective confidential mine will be discussed, to exemplified
the methodology results.

Methodology Results

The groups Massive Berthierite, Fine Arsenopyrite, Banded Iron Formation, Schist and Pyrite, Quartz
and Sulphide, Quartz and Native Gold and Waste (defined by chlorite schist, graphite schist - Figure
6 and 7) represent the lithotype defined. After classification, the electronic signals and the
identification of sources of Au provided by the Bench-Scale Test were used to elaborate the test
program.

7
 Figure 6 Textures and classified images from the XRT sorter

Figure 7 Sources of Gold in different orebodies

The main sources of gold are Berthierite, Fine Arsenopyrite, Quartz and Sulphide. The frequency/
amount of these classes is variable for each ore body section. Using the collected information (XRT
images, density and chemical analyses), a logic for the program was created (Figure 8). Three ore
material zones and one waste material zone were defined by the combination of different signals
from XRT analysis. All grains classified in the three ore zones should be ejected by the machine
producing the new mill feed.

8
 Figure 8 Ore Sorting Program by textures

The Lithotype and Performance Test results by lithotype showed the potential to achieve feasible
gold recovery and where the sources of gold losses are. The waste fractions were collected and hand-
sorting exercises were performed to investigate gold losses showing it is predominantly in the Quartz
and very fine and disseminated Arsenopyrite phases (Figure 9).

9
 Figure 9 Graph showing the results of gold contents of ore sorting waste

To optimize and improve the process, the programs were recalibrated and optimized to avoid the
losses. It is also tanking in consideration the use of another sensor technology to work as a scavenger
in the process, as laser, optical and infrared sensors.
The final gold balance after combining XRT and Laser technologies showed potential to achieve 97.
5% of gold recovery, concentrating in 40% of the mass. The lithological balance demonstrated the
efficiency of the two technologies (XRT and Laser) in selecting the carrier classes of Au in one of the
AngloGold Ashanti sites.

CONCLUSION

Sensor-based ore sorting is a high level disruptive technology that impacts on all sub-processes of
mineral production. It is therefore crucial for the successful analyses of the financial and technical
feasibility, for risk analysis, project execution and operation to involve a team of experienced experts
in an iterative and holistic project (Robben et al, 2015). Standard project development phases include
Bench-Scale testing for Calibration and Validation of the Detection Efficiency, Performance testing
for analysis of the Total Process Efficiency and Piloting for increasing the sample size and for
practicing operational procedures.
AngloGold Ashanti has applied a new and extensive geological approach in the methodology based
on the correlation of geological information and information collected by different sensor-based ore
sorting sensors. The experience applying this methodology in several mines proved how
critical/important the geological approach is to investigate the ore sorting feasibility in complex ores,
in particularly considering gold ores. This is because gold is not detected itself, but correlations
between detectable characteristics and gold content need to be present, well understood and reliable
to understand the gold deportation to product and waste fractions.
The methodology using lithological balance has already proved feasible in cases where a simple
testing approach not considering a strong geological investigation would fail. Potential results
achieving 97. 5% of gold recovery in 40% of the mass proves its success. In addition, the information
acquired supports the stability of other processes such as metallurgy and mine geology.

REFERENCES
Robben, C. (2013). Characteristics of Sensor-Based Sorting Technology and Implementation in
Mining. Aachen, Germany: Shaker Verlag.

10
Robben, C., Takala, A., Rantamaki, M. (2015). Interdisciplinary sensor-based sorting process design
and project development. Aachen, Germany: Proceedings of the Sensor-based sorting and Control
Conference.

GOLDFARB, R. Orogenic gold and geologic time: a global synthesis. Ore Geology Reviews, v. 18, p.
1-75, 2001.

Dunham S. & Vann J. 2007. Geometallurgy geostatistics and project value does you block model tell
you what you need to know? Australasian Institute of mining and metallurgy. In: Project Evaluation
Conference.

Gold Mines, Quadrilátero Ferrífero, Brazil. Economic Geology, v. 96, p. 25-48, 2001.

SALES, M. The Geological Setting of the Lamego Banded Iron-Formation-Hosted Gold Deposit,
Quadrilátero Ferrífero Distric, Minas Gerais-Brazil.1998 182 f. Dissertação (mestrado), Queen's
Universit.

Dalmijn, W.L., Maltha, C., van Houwlingen, J., Frisch, H., 1995. New developments in automated
sorting and automatic quality control systems for glass recycling industry. Proceedings R’95
Recovery Recycling Re-integration, Switzerland, vol. III, pp. 352– 357.

Quay, L., Sattmann, R., Noll, R., 1997. Automatic laser based material identification and marketing,
a new approach. Proceedings Conference on Lasers in Material Processing, Jun. 16–20, Munich,
Germany.

Veerendra, e. S. Mohan Rao Application of image processing and radial basis neural network
techniques for ore sorting and ore classification, Volume 18, Issue 15, December 2005, Pages 1412-
1420

11