Techniques of Pre-Concentration by Sensor-Based Sorting and Froth Flotation Concentration Applied to Sulfide Ores—A Review

Abstract

The use of pre-concentration and optimization of concentration methods have been the focus of the modern mineral industry. Sensor-based sorting equipment and flotation are key players in that movement. This study provides an overview of the main sensor-based sorting techniques and their uses, focusing on sulfides, addressing performance analysis methodologies, and giving the advantages and limitations of the method. An overview of the flotation technique is also presented, covering its basic principles of operation, as well as its main applications in sulfides, its interactions with pre-concentration, and some opportunities and perspectives on the method, such as water reuse impacts, tailing reprocessing, etc. Case studies are presented addressing the influence of the techniques on each other and some future prospects for the mining sector, such as deep-sea mining (DSM) and the use of artificial intelligence (AI).

1. Introduction

The importance of employing pre-concentration methods (especially sensor-based sorting) and the continuous drive to improve flotation processes are the focus of current mineral processing, aligning with cost reduction (both environmental and financial) and ensuring the rational exploitation of mineral deposits. The general trend of declining average ore grades in deposits has made the mining of low-grade ores a reality for many operations worldwide, a subject extensively studied by various authors [1,2,3,4,5,6].

On another front, the economic reuse of old deposits—such as tailings, waste, and low-grade ores—has also been widely considered [7,8]. These materials, generated during the mining of old high-grade deposits, often contain grades comparable to those of natural mineral deposits today. According to estimates by Barros et al. [2], it is projected that by the 2040s, all copper (Cu) deposits in major producing countries (e.g., Chile, USA, Peru, Australia, Mexico, and Indonesia) will have grades below 0.5% Cu. This trend is also observed for other mineral commodities, as demonstrated by Mudd [6], who illustrated the decline in grades for Ag, Cu, Pb, Au, Zn, Ni, U, and diamonds from 1840 to 2006.

Flotation is widely regarded as the most important beneficiation method globally, used in mineral concentration and recycling. During the 20th century, flotation was a revolutionary technique in mineral processing, with significant contributions in research, development, new devices, etc. [9,10,11]. However, despite its current importance, the high energy consumption associated with flotation, particularly in grinding, has spurred the search for alternative methods.

Sensor-based sorting (SBS) technology, including smart equipment utilizing machine learning, has emerged as a key area of research for engineering students and professionals in the field [5]. However, although research using sensor-based sorting has been ongoing for many years, it was only after the early 2000s that this technology gained prominence in the mining industry [12]. Prior to this, processing techniques using fine granulometries like flotation dominated this scenery, as practically the sole route in the mineral processing of sulfides.

Typically, SBS alone cannot achieve the enrichment ratios required for major mineral commodities, primarily due to the limited degree of liberation achievable in the particle size ranges handled by such equipment. Nevertheless, flotation without pre-concentration requires the grinding of high proportions of gangue minerals to particle sizes below 200 µm, a process that generates high demand for mill components and energy. Thus, the combination of pre-concentration and concentration methods appears to be the natural path forward for the modern mineral industry.

Figure 1, adapted from Murphy et al. [13], demonstrates the potential impact of introducing pre-concentration processes into traditional processing routes. In traditional routes, costs tend to increase steadily as ore grades decline. In contrast, routes incorporating pre-concentration are expected to mitigate this cost progression due to the benefits provided by sensor-based sorting techniques.

The objectives of the present study were as follows:

• To conduct a critical review of the main concepts and techniques of froth flotation and pre-concentration by sensor-based sorting;
• To identify potential and limitations in the application of the technique, as well as future perspectives on the subject.

2. Beneficiation of Metallic Sulfides

Metallic sulfides are characterized by the association of a given element with sulfur (Me + S), forming a molecule that often includes more than one type of metal atom. These sulfides are named according to their predominant metal and most industrial metals occur in this form.

2.1. Main Metallic Sulfides and Their Beneficiation

The majority of metallic mineral commodities occur in as sulfides, which are included in the group of industrial metals. The main metals beneficiated from sulfides are Cu, Zn, Pb, and Ni, representing 12%, 7%, 2%, and 1% of the total, respectively, as shown in Figure 2 [14].

Copper is the third most mined industrial metal in the world (Figure 2), accounting for 12% of the total production, with global reserves potentially reaching 3.5 billion tons [14]. Cu occurs in both sulfide and oxide forms, with the majority of production (approximately 80%) coming from sulfides [15]. Its larger mines are located in Chile, Australia, Peru, China, the USA, and Mexico [2], with flotation the most employed method of concentration.

Table 1. Nomenclature and chemical composition of the main Cu minerals.

Type	Mineral	Formula
Sulfides	Chalcocite	Cu2S
	Covellite	CuS
	Chalcopyrite	CuFeS2
	Bornite	Cu5FeS4
	Stannite	Cu2FeSnS4
	Enargite	Cu3AsS4
	Tennantite	Cu12As4S13
	Famatinite	Cu3SbS4
	Tetrahedrite	Cu12Sb4S13
Others	Cuprite	Cu2O
	Tenorite	CuO
	Chrysocolla	(Cu,Al)2H2Si2O5(OH)4·nH2O
	Atacamite	Cu2Cl(OH)3
	Malachite	Cu2CO3(OH)2
	Azurite	Cu3(CO3)2(OH)2

provides the nomenclature and chemical compositions of the main copper minerals.

The most common zinc mineral is the sulfide sphalerite (ZnS), also known as blende. Other zinc minerals include silicates like willemite (Zn₂SiO₄) and hemimorphite (Calamine) (Zn₄Si₂O₇(OH)2H₂O), oxides such as zincite (ZnO) and franklinite (ZnFe₂O₄), and the carbonate smithsonite (ZnCO₃) [16,17]. Zinc beneficiation is typically carried out by the flotation process, although leaching and gravimetric methods are also used.

The primary lead mineral is galena (PbS), a dense sulfide that exhibits a metallic shine [16]. Due to its high density, lead is usually processed using gravity methods for coarse fractions and flotation for fine particles.

As for nickel, it accounts for 2% of industrial metal production, with approximately 40% occurring as sulfides [14]. Nickel is widely used in metal alloys (with Cu, for example), special steels, and technological equipment such as nickel–cadmium batteries and catalysts. The main nickel sulfide is pentlandite (FeNiS), often associated with chalcopyrite or pyrrhotite, and these are concentrated by flotation.

Polymetallic deposits, where multiple economically valuable minerals occur together, are common among sulfides. Classic examples of this type of formation are occurrences of lead–zinc [16], copper–lead [18], copper–lead–zinc [19,20], copper–silver [21], and copper–silver–gold. The beneficiation process for these deposits is typically divided into several stages, where flotation can be used to selectively depress or float certain elements, maximizing metallurgical recoveries.

2.2. Concentration by Flotation

2.2.1. General Flotation Concepts

Flotation is a technique for the separation/concentration of minerals based on the wettability of mineral surfaces, a property that can be selectively altered through the use of chemical reagents. The process involves three key processes in solid (mineral), liquid (water), and gas (air) phases [22]:

• Collision of mineral particles with air bubbles;
• Adhesion (adsorption) and/or formation of air bubble–particle aggregate;
• Transport of the aggregate to the liquid surface, where the particles are collected in a froth format.

Figure 3, adapted from [9,10], illustrates this process.

Metallic sulfides generally exhibit very similar behavior during flotation, enabling operational techniques for their separation from gangue minerals through bulk and/or selective flotation [23]. The performance of flotation is influenced by several key factors, including ore characteristics, mechanical cell parameters (such as cell design and configuration), and operating conditions (e.g., pH, pulp density, and aeration rate) [24,25]. Additionally, the selection and use of reagents (collectors, frothers, and modifiers) play a fundamental role in the success of the flotation process. For further insights into these topics, additional studies provide detailed discussions [9,10,11,23,24,25,26,27,28,29,30,31].

2.2.2. Challenges and Opportunities in Flotation

Michaux et al. [32] proposed a model for simulations of mineral-processing scenarios that rationalize water use by considering the impacts of recycled water on the process. However, the composition of recycled or alternative water can significantly affect flotation performance and plant reliability, necessitating careful consideration of its potential negative impacts [33].

Regarding alternative sources, Castro [34] states that marine or saline waters are often considered detrimental to the flotation of copper and molybdenum ores. However, the effect of sodium chloride and other electrolytes can enhance the floatability of some naturally hydrophobic minerals. Therefore, salinity itself should not be regarded as a limitation to its use in flotation.

Over 60% of the world’s copper deposits are porphyry deposits, e.g., Chilean and North American deposits, which collectively produce over 175 million metric tons [35] annually. However, traditional routes of flotation frequently struggle to process these ores and several deposits like them economically. SBS offers a potential solution by pre-concentrating these ores to a grade suitable for flotation. Cetin et al. [36] precisely studied this situation and obtained promising results for copper pre-concentration using sensor-based sorters equipped with multisensors, and CSIRO [37] presented interesting information as well. Tabelin et al. [35] and Tanaka et al. [38] are other examples of studies on flotation possibilities to process complex ores.

Tailing reprocessing is another emerging opportunity, given the global decline in ore grades. Hence, studies such as those by Santander and Valderrama [8] have explored the recovery of sulfides from old tailings using flotation. Videla et al. [39] determined the positive influence of ultrasound use in the flotation of copper tailings from a mine in Chile, with a copper content of 0.12%. More examples can be found in [40,41,42].

Innovations in flotation technology, such as carrier flotation using glass bubbles, aerosol collectors, and fluidized bed flotation, have shown the potential to enhance performance [43,44,45,46,47,48,49,50,51,52]. These technologies, along with others, confirm the potential for improving the overall efficiency and effectiveness of flotation processes.

3. Pre-Concentration by Sensor-Based Sorting (SBS)

3.1. General SBS Concepts

The process of pre-concentration by SBS serves essentially to enact pre-concentration (primarily for low-grade ores), quality control of final products, and separation of ores into high-grade and low-grade fractions [12]. Among the options, the most commonly employed strategy in the mining industry is the use of sensor-based sorting for pre-concentration. This is typically implemented between the crushing and grinding processes to remove waste rocks, thereby improving the efficiency of downstream processing [53].

According to Ergün et al. [54], sensor-based sorting equipment (ore-sorter) consists of four basic components:

• Feeding presentation system;
• Detection system (sensor/sensors);
• Processing system;
• Material separation system.

The basic layout of such equipment is shown in Figure 4. Additionally, a key feature of SBS equipment is its ability to operate and perform analyses without direct contact with the material, enabling non-destructive and efficient processing [55].

For each of the previous stages, there are specific considerations that need to be made in order to achieve the optimal equipment performance. These considerations involve five basic steps (

Table 2. Five basic steps for optimal sensor-based sorting performance.

Step		Considerations
1	Preparation	- Material may need to be washed, depending on the property being detected. - Key factors: removing surface contaminants, improve sensor accuracy.
2	Feeding	- Preliminary classification based on a specific particle size range (typically maintaining a 3:1 ratio between larger and smaller particles). - Key factors: belt or feeder fill factor, feeding speed, and liberation degree.
3	Presentation to the sensor	- Proper arrangement of the material on the conveyor belt is critical. - Key factors: fill factor of the belt and uniformity of particle distribution, distance between the samples and the sensor, conveyor belt velocity.
4	Sensor detection	- Detectable contrast between particles is essential. - Appropriate sensor selection (or combination of sensors) must align with material properties. - Key factors: sensor resolution, calibration and recalibration, especially to account for deposit variations.
5	Separation	- Considerations include the quality of the air feeding the ejection system and the type of separation device used to ensure precise and efficient material sorting.

):

Another important consideration is the method of material analysis, which can be divided in two categories: bulk ore sorting (BOS) or particle sorting. In BOS separation, a volume of ore is analyzed on the conveyor belt and the whole composition is considered; therefore, the preservation of the natural heterogeneity between ore and waste during feeding is very important to efficient separation. This method is particularly effective for large-scale operations where rapid processing of bulk material is required.

On other hand, the particle sorting is based on the features of each particle individually, demanding that the particles are separated from each other and presented to the sensors in just one layer on the conveyor belt. This method allows for highly precise separation, as each particle is analyzed and sorted independently, making it particularly useful for high-value ores or when a high degree of accuracy is required.

3.2. Techniques and Equipment

The types of features detectable by sensors in sensor-based sorting include surface features, whole-particle features, secondary features (such as color, reflection, brightness, conductivity, density, magnetic susceptibility, etc.), and primary properties (mineralogical, chemical, and physical composition) [12].

Table 3. Techniques of sensor-based sorting applied to sulfide ores, adapted from [12,56].

Spectrum	Detected Features	Penetration	Interaction with	Examples of Applications
X-ray transmission (XRT)	Primary features, atomic density	Deep	Transmission of RX through the material	Metals, precious metals, industrial minerals, coal, diamonds, recycling
Visible light (VIS)	Secondary features	Superficial	Reflection, absorption, transmission, luminescence	Industrial minerals, precious stones, recycling, diamonds
Near-infrared (NIR)	Secondary features	Superficial	Monochromatic reflection and absorption	Base metal ores, industrial minerals, precious stones, diamonds
Infrared (IR) + microwaves	Secondary features	Superficial	Heat dissipation after microwave submission	Metals, industrial minerals
Radiowaves [magnetic resonance (MR)]	Primary features, mineralogy	Deep	Excitation and detection of spectral radiowave lines	Bulk ore sorting (BOS), calcopyrite
Alternating current (AC)	Secondary features	Deep	Conductivity, magnetic susceptibility	Iron and other base metals, recycling

provides an overview of the main sensor types applied to sulfide selection [12,56].

3.2.1. X-Ray Sensors

X-ray transmission (XRT) sensors are arguably the most versatile type of sensor, finding widespread use in applications where a measurable density difference exists between ore and gangue or between two types of ore [13]. This atomic density difference attenuates radiation differently, either proportionally (using single-energy methods) or non-proportionally (using dual-energy methods), based on the thickness of the specimens [57].

The effect of thickness, as described by [58], is as significant as variations in particle thickness. In other words, if the range of these differences is as large as the density variation, simple energy sensors are not suitable. As noted by Strauss [57], this issue can be mitigated through conditioning, such as using classifiers to limit the range between the largest and smallest particles to a ratio of approximately 1:3.

With dual-energy sensors, which are widely used, radiation is generated in two bands, allowing each pixel to be classified according to its specific atomic density, attenuating the effect of thickness. The DE-XRT sensor system records the interaction of radiation with the material through two channels of different radiation intensities (high and low energies). Figure 5 provides a complete illustration of this phenomenon.

This process generates grayscale images, where dark colors represent denser areas and light colors represent less dense areas. Specialized software can then process these images, producing a single image with a hierarchical color scale (false color) for enhanced analysis [59].

3.2.2. Visible Light

Visual sensor technology (or VISs—visual image sensors) operates within the wavelength range of the human eye’s vision sensitivity (390–780 nm). These sensors possess a photosensitive area that, upon interacting with varying light intensities reflected by a specific object, generates measurable electrical signals. Devices that work in conjunction with optical systems are commonly referred to as digital cameras [60].

These cameras utilize a component known as a charge-coupled device (CCD), which functions similarly to photographic film. Additionally, cameras equipped with complementary metal-oxide semiconductors (CMOSs) are now widely used as sensors. This system adopts the same technology as other typical semiconductors, resulting in reduced costs [61].

The Figure 6 presents the scheme of operation, where the light reflected from the sample is captured by the CCD camera, transformed in digital data and sent to a processing unit. The images generated by these sensors are composed using a color model, with RGB (red, green, and blue) being the most commonly used. Thus, each color can assume 28 or 256 different intensity values, ranging from 0 to 255.

These sensors have a wide range of applications in sensor-based sorting [54]; depending on the nature of the deposit, optical classification can be used for pre-concentration or to obtain a high-quality final product for various minerals, such as magnesite, limestone, quartz, feldspar, coal, barite, or copper. Another important consideration, according to Kolacz [62], is the conditioning of the material before feeding it to the equipment. The material should be as free as possible from impurities on its surface. While washing is the most traditional method for cleaning these materials, it can pose challenges if there are restrictions on moisture in subsequent processes or if the material cannot be wetted due to intrinsic reasons or water scarcity at the deposit site.

3.2.3. Laser

Three-dimensional laser sensors provide information about surface characteristics, including shape, size, roughness, and brightness. When used in conjunction with other sensors, they allow for precise determination of the particle’s exact position on the conveyor belt, facilitating accurate ejection. In summary, their operation relies on the detection of laser reflection and diffraction from the surface of the analyzed particles [57].

LIBS (laser-induced breakdown spectroscopy) is a detection system that considers the elemental composition of the material being analyzed. It generates plasma when a pulsed laser interacts with the material, with responses potentially reaching temperatures of 10,000 K. The photons emitted by the plasma are captured by a spectrophotometer that can be calibrated across different spectral ranges [63].

For laser-based sensor sorting equipment, separation is based on the penetration of light according to the material’s structure. Strauss [57] notes that some materials lacking sufficient contrast for color-based separation but exhibiting compositional differences can respond well to laser use. Figure 7 illustrates an example of detecting a quartz-rich area in a rock sample, which in this case hosts gold mineralization. Further details about the technique can be found in [60,63,64,65].

3.2.4. Near-Infrared and Short-Wave Infrared (VNIR-SWIR)

Sensor-based sorting systems that operate using near-infrared (NIR) spectroscopy analyze the interaction between electromagnetic radiation and rock [66]. According to Robben and Wotruba [67], the advantage of this type of sensor is that it is based on the primary properties of minerals. Each mineral species emits a different spectral response when exposed to NIR, allowing for precise identification and separation.

In operational terms, the equipment exposes the material to a source of energy in the NIR range. Some of the radiation is absorbed by the mineral molecules, while another portion is reflected and detected, resulting in a specific spectral signature that is recorded [67]. These devices can operate over a wider particle size range since the analysis considers the chemical composition of the ore and is independent of thickness or other physical features.

VNIR-SWIR hyperspectral sensors work in the wavelength range of 380–2500 nm and have been widely used in material identification and characterization across various scientific fields [68,69]. Regarding their use specifically in the SBS field, studies such as [70,71,72] are examples of the successful use of these sensors on metallic minerals.

3.2.5. Microwaves (MW-IRT)

Microwave (MW) heating has been proposed as a selection technique based on the differential elevation of temperatures of ore particles, enabling the distinction between different grades and/or compositions. This method relies on the dielectric properties of the material, which is linked to its ability to heat to varying degrees when exposed to MW heating. Additionally, consequent changes in the electromagnetic field can be measured to further differentiate materials [73,74].

This mechanism is highly effective for elements that strongly absorb microwaves (e.g., nickel, copper, gold, and other metallic sulfides) as well as minerals containing molecular or free water (e.g., smectite clays, which may also have undesired effects) [75]. These minerals can be separated from those that exhibit little or no response to MWs, such as quartz, feldspar, mica, and other non-sulfide minerals [73]. Examples of this differential heating can be observed in

Table 4. Heating responses when using the microwave sensor for various mineral groups, adapted from [64].

Category	Mineral	Temp. (°C)	Time (min)
Easy heating	FeS2	1019	6.75
	PbS	956	7.00
	CuFeS2	920	1.00
Hard heating	SiO2	79	7.00
	Al2O3	78	4.50
	KAlSi3O8	67	7.00
	CaCO3	74	4.25

, which was adapted from [64].

The detection of this sample heating is performed by thermal image sensors (infrared thermal—IRT), which collect information immediately after exposure to microwaves (MWs). In the IR spectrum, the heating can be observed on the samples surface and differentiated from other samples using a false color [64,74]. Figure 8 presents a group of copper ore samples analyzed in laboratory, where the response under a MW source could be determined by a thermal scan, showing hot zones (red/yellow) coinciding with high-grade particles and cold zones (green/blue) coinciding with low-grade or non-mineralized particles.

3.2.6. Radiowaves

This technique is capable of detecting the mineral signature based on the radiofrequency pulses that are returned by each one, also known as magnetic resonance (MR). Due to the inherent resonance difference between two distinct material types, the echo magnitude is proportional to the number of crystalline units of a specific mineral, varying according to its composition/concentration, as shown in

Table 5. Response to radiowaves by mineral type.

Mineral	Category	MR Sensitivity
Chalcopyrite	Copper	High
Cubanite	Copper	High
Covellite	Copper	Medium
Chalcocite	Copper	Medium
Enargite	Copper	Low
Tennantite	Copper	Low
Cuprite + delafossites	Copper	High
Tenorite	Copper	Low
Arsenopyrite	Arsenic	High
Orpiment	Arsenic	High
Realgar	Arsenic	High
Lollingite	Arsenic	High
Niccolite	Nickel/Arsenic	Medium
Hematite	Iron	High
Magnetite	Iron	Very High
Maghemite	Iron	High
Pyrrhotite	Iron	High
Bismuthinite + others	Several	Medium
Stibnite + others	Several	High
Zircon	Zircon	Low
Cobaltite	Cobalt	High

, which was adapted from [76].

The most relevant works related to radiofrequency wave resonance (RF) are associated with Australia’s National Science Agency [37], which has proposed the use of this technique as a potential revolution in the sensor-based sorting sector. Because it is based on the compositional characteristics of minerals, this technique can easily discriminate between different types of ores and gangues without the need for recalibration. However, according to [36], there are also some limitations, such as the inability of the technique to detect certain mineral species, for example, bornite (one of the most abundant Cu sulfides).

Ferrari-John et al. [75] mention another application of RF waves, used for differential particle heating. According to the authors, this method can yield better results than using microwaves (MWs), avoiding the potential noise generated by hydrated or free-water minerals. Further details on this application can be found in their paper.

3.2.7. Electromagnetic Devices

In recent years, ore sorters based on electromagnetic methods have been rapidly developed, showing significant potential for pre-concentration, particularly due to their possible use in small particle sizes. Two approaches are feasible for this separation: one exploits the galvanic characteristics of particles, while the other focuses on inductive properties [77].

The galvanic method has several limitations, which can be found in the literature. However, the inductive method, which measures rock impedance based on Faraday’s law, is highly applicable. In this method (Figure 9), a pair of electrodes is installed beneath the conveyor belt (coil system) generating a magnetic field and variations in voltage when rocks are subjected to this field are measured (processing) and used as separation criteria [77].

Magnetic sorting is also considered a type of ore sorter; however, its reliance on a magnetic roll and the absence of advanced sensor systems raise questions about its classification as a true sensor-based sorting method. As a result, these devices will not be addressed in this paper. For those interested in further details, references [78,79] can be consulted.

3.2.8. Sensor Fusion

In addition to sensors that inherently require a combination of devices for their operation, such as MW + IR or laser + camera, various complex mineral materials only exhibit a good response to sensor-based sorting when multiple sensors are used simultaneously. In such cases, one sensor can compensate for any limitations of the other. Kattentidt et al. [80] asserts that the use of equipment combining multiple sensors holds significant potential for gains in the processes where they are employed. This potential is realized in the sensor-based sorting equipment currently manufactured by major industry players.

The concept of technical synergy, as reported by Cetin et al. [36], confirms that sensor fusion techniques provide higher precision for grade measurements. The authors cite an example of using MR (magnetic resonance) in conjunction with PGNAA (prompt gamma neutron activation analysis) for low-grade copper concentration. Individually, each sensor may lack economic viability or exhibit very low performance [net smelter return (NSR) close to zero or below]. However, when both techniques are used in combination, the phenomenon of “technical synergy” emerges, resulting in a positive net smelter return (NSR) in all simulated scenarios. Both these cases are presented in Figure 10.

3.2.9. Other Techniques

In addition to the previously mentioned techniques,

Table 6. Other sensor-based sorting techniques commonly used for non-sulfide ores.

Spectrum	Detected Features	Penetration	Interaction with	Examples of Applications
Gamma-radiation	Secondary features, emission	Deep	Natural gamma radiation	Uranium and other radioactive minerals, precious metals
X-ray fluorescence (XRF)	Secondary features, emission	Superficial	Electrons of external atomic layer	Diamonds, sample analysis
X-ray luminescence (XRL)	Secondary features	Superficial	Excitation of luminescence by X-rays	Diamonds
Ultraviolet (UV)	Secondary features	Superficial	Reflection, absorption, transmission, luminescence	Diamonds

provides examples of other sensors that are predominantly employed for different types of ores, such as diamonds. Works such as [56,64,81,82,83,84,85,86,87] serve as valuable sources for further details.

3.3. Technical Challenges

The primary criteria for successfully applying the SBS technique are threefold [64,88]:

• Existence of a sufficient degree of ore vs. gangue liberation;
• Feasibility of identifying ore and gangue using an available SBS technique;
• Production rate.

Regarding the first criterion, the degree of liberation is directly linked to the material’s particle size distribution. In other words, the smaller the particle size, the greater the likelihood that ore minerals are separated from gangue minerals. However, operating SBS in excessively small size ranges becomes a significant challenge, as equipment productivity drops substantially and comminution costs rise. For instance, the capacity of a DE-XRT device decreases from 110 t/h·m in a 40–75 mm range to 15 t/h·m in a 6–8 mm range [88].

Several studies indicate that better performance results are not necessarily achieved in smaller particle size ranges, even though a higher degree of liberation is expected. Dias Júnior [89] conducted tests on chromite ore, analyzing recovery and grades through concentration using a DE-XRT sensor and neural network simulation. In both cases, the highest metallurgical recoveries were observed in the coarser size fraction (+4–5″), with 91.56% and 54.59%, respectively. These values dropped to 18.49% and 31.88% in the intermediate fraction (+3–4″) and to 1.91% and 13.33% in the finest fraction (+2–3″). Amorim Júnior [90] tested DE-XRT sensors applied to zinc ore, obtaining metallurgical recoveries ranging from 45% to 90% for the finer fraction (+16–35 mm) and from 45% to 97% for the coarser fraction (+35–100 mm).

Building on the work of Peukert et al. [91] and Bamber [3], Figure 11 illustrates the distribution of metallurgical recovery versus particle size for 49 results from several studies, such as [92,93,94,95,96,97,98,99] (the complete dataset is available in the Supplementary Materials, Table S1). Notably, all metallurgical recoveries below 85% fall within the smaller particle size ranges (i.e., particle sizes below 110 mm), as delineated by the red lines.

Sensor resolution is also a key factor, as smaller particle sizes result in lower representativity in terms of pixel count or sensor response, eventually reaching the operational limit of the equipment. Therefore, the optimal particle size range for each SBS operation should be determined by comparing the responses of different size fractions and their respective performance during separation, aiming for the best possible balance between the degree of liberation and the characteristics of the sensor used.

Further analyzing the database used to construct the previous graph (Figure 11) reveals notable trends, as shown in the histograms in Figure 12. Specifically, 75% of the analyzed results exhibited metallurgical recoveries above 85%. In terms of feed particle size, the most commonly used range was between 98 and 128 mm. However, it is also evident that more than 60% of the studies employed particle size ranges below 98 mm, which tends to result in lower recoveries.

Regarding the second criterion, Section 3.2 provided a detailed overview of the most commonly used SBS techniques and their affinities with various types of sulfide mineral commodities. This information can serve as a reference for determining the feasibility of identifying ore and gangue using available SBS techniques.

As the third criterion, the production rate, referring to the capacity per meter of belt width for SBS equipment (t/h×m), as parametrized by Robben et al. [88], is presented in

Table 7. Number of the 49 studied cases classified within each range of production capacity in SBS equipment. Adapted from [88].

Particle Size Interval (mm)	Capacity Limit (t/h×m)	No. of Cases	%
5.6–8	15	1	2.04%
8–20	30	1	2.04%
20–40	60	13	26.53%
>40	>110	34	69.39%
		49	100.00%

. The results, simulated using data from the previous 49 studied cases, show that approximately 30% of these cases had a production limit of 60 t/h×m. In practical terms, achieving a production rate of 180 t/h would require at least three ore sorters with a 1 m width. However, this rate is still considered low for large-scale mining operations.

3.4. Performance Analysis

Traditional performance indicators used in beneficiation processes are also applied to ore sorters. These include the enrichment ratio, mass recovery, metallurgical recovery, economic indicators, and more [36,89,90,100,101,102,103,104]. When it comes to analyzing the performance of separators, relevant concepts for graphical representation of these indices are detailed in the publications by [105,106]. These studies demonstrate the possibility of using techniques already employed in other binary classifications for a more detailed performance analysis of sensor-based sorting [107].

Among these indices are accuracy, precision, specificity, sensitivity, true/false positives/negatives, and others. These indicators are related to the so-called confusion matrix (

Table 8. Example of a confusion matrix, adapted from [105,106].

Feed	Product	Waste
Positive fraction	True positives (TPs)	False negatives (FNs)
Negative fraction	False positives (FPs)	True negatives (TNs)
∑	Product fraction (P)	Waste fraction (N)

), which represents the mass balance obtained during separation. The formulas for calculating these indices are given by Equations (1)–(3).

Accuracy = (TP + TN)/(P + N),

(1)

Specificity = TN/(FP + TN) = TN/N = 1−FPr

(2)

Sensitivity = TP/(FN + TP) = TP/P = TPr

(3)

When speaking about specificity and sensitivity indices, it is worth noting that these are the indicators used in constructing the so-called ROC (receiver operating characteristics) curve. The ROC curve is plotted with sensitivity values ranging from 0 to 1 on the y-axis, representing the true positive rate (TPr) (Figure 13). Meanwhile, the x-axis represents the false positive rate (FPr), obtained by subtracting 1 from specificity [107,108,109].

These indicators were employed in the performance analysis of the separation of magnesite, quartz, lignite, hematite, copper ore, and gold ore samples [103]. They simulated separation scenarios across various particle size ranges, ranging from −50 + 9.5 mm to −150 + 18 mm, while also considering variations in the hourly feed tonnage.

The authors’ main conclusions were as follows:

• Samples with higher ore vs. reject contrast (e.g., magnesite and quartz) yielded better results than samples with moderate contrast (lignite and hematite) or low contrast (copper and gold).
• Efficiency of classification decreased as the feed rate increased.

This method of performance analysis, or sortability, has also been applied by other researchers [36,73,74,108,109].

4. Impacts of Pre-Concentration in the Beneficiation Process

Given the lack of direct and explicit studies on the impacts of pre-concentration in the mineral beneficiation process, substantial contributions in this area remain scarce [110]. The studies presented in this topic underscore the need for further research to address that gap.

4.1. Case Study I: QZ Ohio Ore—Australia

A study conducted by Rizmanoski [74] in Australia tested the applicability of microwave (MW) technology for separating copper ore known as “QZ Ohio Ore”. The study analyzed the impact of this separation on subsequent grinding and flotation stages. The ore was divided into three cut-off grades based on its response to MWs (approximately 2 kg each): a hot fraction (highest sensor response), an intermediate fraction, and a cold fraction (lowest response).

As shown in Figure 14A, for the two tested grinding times, the P75 particle size distribution exhibited a significant reduction, indicating a positive influence of pre-concentration on grinding efficiency. When considering the 15 min grinding curve, the particle size reduction was approximately 15%, while for the first 10 min, it was around 17%.

There was also a similarly significant contribution to the flotation results, showing an increase in mass recovery for the “hot” fraction, which was approximately 30% higher at all times compared to the other fractions. The graph presented in Figure 14B displays cumulative Cu recovery data over flotation time, where the curve representing the hot material yielded the best results within the first minute of flotation, indicating better kinetics in practice and requiring less residence time in the tanks for metal recovery.

4.2. Case Study II: Polymetallic Ore—Aripuanã, Brazil

A case study conducted by Lopes at al. [111] investigated the application of pre-concentration using sensor-based sorting for polymetallic ore (Zn, Pb, Cu). Approximately 20 tons of samples were processed using XRT sensors. The results for Zn and Pb are demonstrated in Figure 15. The pre-concentration circuit increases the overall metallurgical recovery of zinc and lead, by 2% and 11%, respectively. The mass reductions in the total ROM almost equal 40% and, additionally, there were gains in all of other aspects analyzed in the plant mass balance. Notably, the Pb fraction exhibited more significant improvements, with a reduction in losses from 15.9 tons per day (tpd) to 9.3 tpd and with recovery increasing from 68.8 to nearly 80.0 tpd.

It is also worth noting that the study considered reducing the plant’s feed mass while keeping the initial feed constant. However, if there was an increase in feed from the mine, it would be possible to adjust the pre-concentration product to achieve roughly the same current initial feed (but with a higher grade), thereby increasing daily metal production.

4.3. Case Study III: Córrego do Sítio Mine—Brazil

A study conducted by Assis et al. [112] compared the performance of two sensor-based sorting techniques, namely XRT and laser, applied to a gold ore from the Córrego do Sítio Mine in the state of Minas Gerais, Brazil. Figure 16 demonstrates the results obtained with each sensor, highlighting that the best recoveries were achieved when both techniques were used simultaneously. Notably, this approach resulted in good material removal and higher recovery of the target ore (Au).

When considering the two techniques separately, it can be asserted that the laser achieved a superior performance. Specifically, it recovered more gold (72%) from a smaller mass (27% of the feed), although sulfur (S) recovery was lower. In contrast, the XRT sensor recovered only 57% of the gold, while recovering a bigger amount of the mass (32%).

This same study also tested gravity separation and flotation as subsequent concentration processes. When focusing on flotation, it achieved gold recovery rates of up to 89.55%, with an average of 87.41% across three tests. Unfortunately, the study did not provide data on the current flotation process without pre-concentration, making a direct comparison impossible.

4.4. Case Study IV: Phu Kam Mine—Laos

An interesting study conducted by Reple [113] utilized copper ore from the Phu Kam mine, analyzing the impact of bulk ore sorting (BOS) through computational simulations. The study considered both upstream (resources and reserves) and downstream (beneficiation) effects. Figure 17 illustrates the reserve gains for the same grade with the inclusion of BOS (orange curve) versus without BOS (blue curve). For instance, at a copper grade of 0.8% Cu, the reserves increase from 5000 kilotons (kt) to approximately 7500 kt when the SBS equipment is implemented.

According to data, in one of the scenarios, it would be possible to increase the feed grade of the plant from 0.47% Cu to 0.61%, while keeping the mass of fed ROM and produced metal practically unchanged, resulting in approximately 5% profitability gains in the cash flow. On the other hand, if capacity was installed to increase the feed, profitability gains could reach 62%, with plant production going from 63 Kt to 89 Kt.

4.5. Case Study V: San Rafael Tin Mine—Peru

Robben et al. [88] analyzed the SBS system implanted on the San Rafael mine, the largest underground tin mine operating in the world. The SBS device has operated since 2016 and is based on XRT sensor technology, equipped with four ore sorters that work in parallel, with the feed flow divided in four ranges of size: 6–11 mm, 11–22 mm, 22–38 mm, and 38–70 mm.

As the main advantages from this application, the author pointed out six improvements:

• Value addition, through the possibility of treating the material below the cut-off grade (0.9% Sn). In other words, the SBS plant is fed with ore within 0.2%–1.1% Sn, allowing recovery until old marginal piles and generating a pre-concentrated plant feed with 1.9% Sn;
• The plant capacity was increased by 105 t/d, from 2950 to 3200 t/d;
• The plant metal recovery was increased from 90.5 before SBS to 92.5% after it;
• There was an increase in ore reserves, once the SBS feed was composed of 24% low-grade ore, contributing to the overall reserve tonnage and the mine life-span;
• The potential of acid mine drainage generation was reduced, because old stock piles could be recovered after the SBS implantation;
• Another important environmental feature, the tailing disposal, and consequently, the tailing storage facilities, was importantly reduced, replacing it with cheaper and safer dump piles.

4.6. Case Study VI: Souzmetallresource (SMR) Molybdenum Mines—Russia

This study conducted by Lessard et al. [114] aimed to achieve its objectives at two mine sites, called Mine A and Mine B, both mineralized by the main same ore: molybdenite (MoS₂). The work sought to reduce the power consumption and increase the ROM production by applying SBS DE-XRT equipment and making improvements in the plant design model.

For both Mines A and B, the work tested three different thresholds, called relaxed, moderate, and aggressive, where the first is focused on higher concentration factors and the last on higher metal recoveries (

Table 9. Results from SBS tests in Souzmetallresource mines, adapted from [114].

		Mine A	Mine B
Relaxed threshold	Metal recovery	86.4%	65.5%
	Concentration factor	8.8	19.8
	Waste fraction	90.0%	93.7%
Moderate threshold	Metal recovery	88.0%	79.7%
	Concentration factor	5.1	10.2
	Waste fraction	82.9%	79.2%
Aggressive threshold	Metal recovery	93.4%	87.0%
	Concentration factor	2.9	6.0
	Waste fraction	66.5%	54.8%

).

With regard to the reduction in power consumption, the authors pointed out that if the company sustained the ROM feed at the same rate as before SBS technology (3455 t/h), the reduction in gangue material contents in comminution feeding could reduce the total crushing power demand from 5.9 MW to 3.2 MW and the milling demand from 51.0 MW to 16 MW. Using the example of Mine A, the global investment to produce Mo could be reduced from 32,951 kWh/t Mo to 12,775 kWh/t Mo.

Despite the hypothesis above, commonly, companies prefer to keep the plant feed and increase the ROM, offering to their plants a higher-grade ore and more global advantages. Accordingly, the authors presented the possibility of Mine A reducing energy costs from 32,951 kW/t Mo to 12,789 kWh/t Mo. That was possible thanks to the increase in Mo grade at the flotation plant from 0.05% to 0.138%. However, the ROM feed necessarily increased from 3455 t/h to 10,631 t/h. In Mine B, the benefits were the same, with a power consume reduction from 32,951 to 14,606 kWh/t Mo, an increase in Mo grade for the flotation plant from 0.050 to 0.125, and a need to increase the ROM feeding from 3455 to 13,340 t/h.

5. Summary of Sensor-Based Sorting

5.1. Main Advantages and Limitations

Figure 18 illustrates twelve aspects related to the implementation of sensor-based sorting systems in a mining operation, and how sensor-based sorting (SBS) systems can interact with various mine and plant processes, both directly and indirectly. The paragraphs below explore these aspects in accordance with each of these numbers, beginning with advantages.

The proximity to mining faces (1) is an important aspect, as sensor-based sorting (SBS) systems can be deployed near mining fronts, reducing transportation costs and optimizing deposit utilization. Bulk ore sorting (BOS), for example, needs to operate using the natural heterogeneity of the ore, which is higher close to the mine than after several charge and discharge processes. As a result, the pre-concentrated run-of-mine (ROM) material is directly fed to the primary crusher (2), reducing gangue crushing costs and ensuring a more consistent feed grade for the plant. For instance, the energy demand from SBS equipment is around 1–3 kWh/t, while in a milling system, it is around 12–15 kWh/t [64]. Also, BOS can be considered near to the plant, removing gangue before comminution processes and allowing the pre-concentration of ore from more than one mine.

The use of SBS post-primary crushing (3) is the most widely implemented system, even at coarse granulometry. It is typically applied across multiple particle size ranges, removing a significant portion of gangue, reducing grinding and processing costs, and ensuring a more consistent feed grade for the concentration processes. It can also split the plant feed into two streams (high- and low-grade), directing high-grade material to flotation and low-grade material to leaching, for example. Also, the use of SBS in quality control for final products (4) offers a possible application [64], ensuring compliance with customer requirements.

The (5) and (6) arrows show the possibility of using SBS to recover old waste piles (dumps); these can be reprocessed directly from deposits by BOS, if it offers reliable intrinsic heterogeneity, or by particle sorting using crusher facilities. This recovery can contribute largely to reducing acid mine drainage (ADM) generation and, moreover, can generate a new source of ore and expand the total metal resources.

Regarding plant emissions (7), optimizing its operation using SBS can reduce gas emissions and dust from mineral processing, as well as providing the opportunity to process the same amount of metal with less feed tonnage. Moreover, the demand for inputs, such as electricity, water, fuels, chemical reagents, and grinding components, can be reduce importantly, as the work index (WI) of the pre-concentrated material is often lower.

A decrease in tailing generation (9), including pulp and dry tailings, leads to a smaller environmental footprint, demands smaller tailing storage facilities (TSFs), and allows an easier process when obtaining governmental and populational licenses. Another positive impact can be achieved at the mine, with reduced ore losses (12). Furthermore, a lower cut-off grade can be expected (11) and a significant increase in the mineable reserves (10), maximizing the recovery of low-grade ores and/or utilizing marginal or diluted ore zones.

Regarding limitations, the technique can face some difficulty, starting from remote deposits lacking access to energy and other technological resources. Moreover, intrinsic ore features, such as dissemination with low heterogeneity, can also limit SBS use, due to losses from discarding batches with high-grade individual particles (FN) and consequently economic infeasibility. The need to remove fines, retaining a maximum particle range of 3:1 within each size fraction, is also a limitation, because the crusher discharge needs to be divided into three or more flows in multiple deck screens. This limitation leads to a demand to use several parallel SBS devices. The required degree of liberation and the maximum top-size of ROM material can also negatively impact crusher production.

The nature is normally very irregular, mainly with regard to ore deposits, but in some cases, the homogeneity of resources can become a limitation to SBS application, which is expected in disseminated and low-grade deposits. In the same way, old waste dumps that could present conditions for recovery, according to the exploitation features, can also presents low heterogeneity and, consequently, low SBS performances.

5.2. Future Perspectives

There are indications that installing pre-concentration equipment using sensor-based sorting or gravity methods underground could be an interesting alternative for underground mines [13]. Considering the costs associated with transporting material from underground to surface beneficiation plants, where traditional beneficiation processes take place, the possibility of pre-discarding gangue minerals (or a significant portion of them) near the mining faces can be highly advantageous. The following points highlight its benefits:

• Reduction in overall energy consumption (electricity, fuels, etc.);
• Potential use of hydraulic transport systems (hydraulic hoisting);
• Feasibility of more productive bulk mining techniques, as greater ore dilution during extraction becomes viable;
• Opportunity to employ backfill techniques to both optimize mining and prevent future environmental subsidence issues.

Another similar hypothesis is presented by Li et al. [115], exploring the heterogeneity of block caving mining systems, particularly in the edge zones of ore bodies, with similar benefits. The authors cite a system implemented at the Kiruna mine in Sweden, where sensor-based sorting systems examine the load transported by LHDs or underground transport trains. These systems can serve as both online control systems and decision-makers for discarding or utilization (similar to BOS). Other similar studies are [116,117].

The use of artificial intelligence (AI) has been a trend in the sensor-based sorting sector, mainly regarding the ability to generate real-time data and how it can be used in modern sampling systems (much faster and more efficient than traditional systems) [65]. Therefore, several studies are underway that utilize machine learning techniques to optimize sensor-based sorting processes.

Shatwell et al. [53] used a database generated from RGB images and the textures of ore samples containing gold and silver, which were described by experts. They achieved an over 90% correlation in rock sample classification using machine learning models, with a delay of 19 to 44 milliseconds (ms). Other similar studies can be found in [103,118,119].

Following the path of significant technological advancements in intelligent ore sorting, as highlighted by [5], the proposed Autonomous Sorter System (ASS) appears to be the most novel and revolutionary development in sensor-based sorting as of today [120,121]. These authors advocate for the use of AI-powered equipment with multiple sensors, known as the Cyber-Physical Sorting System (CPSS), which can be adapted to various industry segments, including recycling, construction and demolition, mining, and urban mining.

The equipment (Figure 19) consists of a conveyor belt feeding the system, a bank of visual multisensors (hyper-spectral, industrial, and short-wave infrared) capturing images, a robust processing system, and a robotic arm (or multiple rows of laterally arranged air nozzles) separating items into various possible destinations.

6. Conclusions

After reading and analyzing over 150 scientific articles and publications, this study provides a comprehensive review of the current state of pre-concentration by sensor-based sorting and flotation, with a primary focus on sulfides. Regarding sensor-based pre-concentration, it can be stated that among the ten detection techniques presented, X-ray sensors have found the most applicability in sulfide minerals. Notably, combinations of different devices (e.g., laser or PGNA) demonstrate significant performance enhancements.

In terms of performance analysis, until the mid-2010s, publications rarely addressed specific analytical methods, which were primarily limited to enrichment ratios and other indicators commonly used in beneficiation processes. However, there is now a growing use of methods involving confusion matrices and ROC curves, providing a more comprehensive analysis of equipment efficiency.

Regarding the flotation process, the current trend in sulfide flotation is to develop methods that enable operation at low grades and/or coarser particle sizes. This trend aligns with the global decrease in ore grades for all metals, making economic exploitation more challenging. Consequently, numerous studies are attempting to improve and innovate the flotation process. However, most research revisits strategies tested in the past, seeking to determine their contemporary applicability. For example, fluidized bed flotation has been studied since at least 1984 [52].

Another interesting conclusion is the effectiveness of combining techniques (referred to as “technical synergy” by [36]). Examples include using different collectors in flotation, employing multiple sensors simultaneously in sensor-based pre-concentration, and the benefits of combining pre-concentration and concentration techniques. The latter pair, although less common until a few years ago, has become a trend in the current mineral industry, especially with the growth and popularization of sensor-based sorting systems. However, there is still a scarcity of studies quantitatively correlating the impacts of these processes.

In summary, we are currently experiencing an unprecedented period of research and development opportunities, revolutionizing various industries (particularly through artificial intelligence). This progress is already leading to innovative recycling and mining methods, such as deep-sea mining (DSM) [122,123], urban mining, and (why not?) space mining, which already appears as a future trend in publications such as [124,125].