We need to automate our ore-grading process. What's the best AI-powered sensor and ore sorting technology?

Automating ore-grading and sorting relies on combining high-speed multi-sensor hardware with deep learning/AI software capable of per-particle analysis in real time.im-mining.com+2

The industry standard centers around a few powerhouse technology ecosystems, each tailored to different mineralogy and sorting challenges:

• TOMRA Mining (GAINnext & CONTAIN/OBTAIN Ecosystem) : TOMRA leads the AI space with deep learning solutions explicitly trained on convolutional neural networks for mining. Their CONTAIN technology uses real-time X-ray imagery to detect complex, subsurface inclusion-type ores (like nickel, tungsten, and tin) that traditional dual-energy XRT or optical sensors miss. Coupled with OBTAIN for particle-cluster segmentation, it handles complex geometries and dynamically adjusts grade-recovery thresholds on the fly. Check out an overview of their platform on TOMRA Mining.im-mining.com+2
• STEINERT (KSS EVO 6.0 Multi-Sensor Systems) : STEINERT combines 3D laser, color optical cameras, Near-Infrared (NIR) hyperspectral imaging (HSI), and X-ray transmission (XRT). Their edge lies in intelligent mechanical and digital declustering —sorting over a thousand particles per second while preventing rock-clustering misclassifications. It is heavily deployed for early waste rejection, low-grade stockpile upgrading, and complex industrial minerals. Explore application details via STEINERT Solutions.steinertglobal.com+2
• XRF and Laser-Based AI Hybrids : Emerging systems integrate X-ray Fluorescence (XRF) with machine vision to read elemental surface composition mapped via AI models. This approach allows operations to bypass dense media separation (DMS) earlier in the circuit, drastically cutting down crushing and energy costs.youtube.com

Key Sensor Types to Pair with AI

1. X-Ray Transmission (XRT): Best for atomic density separation (e.g., separating barren waste rock from metal-bearing ore).
2. Hyperspectral Imaging (HSI) & NIR: Best for surface mineralogy, color, and texture identification. 3D Laser / Optical: Best for physical shape, volume, and clustering separation prior to ejector firing.steinertglobal.com+1

To narrow down the best setup for your site, could you share:

• What commodity or mineral are you processing (e.g., gold, lithium/spodumene, base metals)?
• What is the current feed size / throughput you need to handle?

Automating ore grading and sorting with AI relies heavily on marrying high-throughput hardware sensors (X-ray, optical, electromagnetic, or laser) with deep learning neural networks . Because "the best" system depends entirely on your specific mineralogy and throughput demands, the top-tier technology ecosystems break down into industry-leading standards:mining-technology.com+2

Leading AI-Powered Ore Sorting Technologies

• TOMRA Mining (OBTAIN™ and CONTAIN™ Ecosystem): Widely regarded as a market leader in integrating deep learning into sensor-based sorting.
  • OBTAIN™ uses convolutional neural networks for single-particle precision in high-throughput sorting, accurately classifying clustered or overlapping rocks independently of belt capacity.
  • CONTAIN™ is purpose-built for complex inclusion-type ores, leveraging real-time X-ray imagery analysis to catch visual and structural patterns that standard automated algorithms miss. You can explore their hardware-software lineup on the TOMRA Mining Technologies portal.im-mining.com+2
• Multispectral & 3D Laser Fusion Systems: Emerging machine-vision architectures combine 3D laser scanners, high-resolution optical multispectral cameras, and near-infrared (NIR) sensors. These systems feed a unified data stream into edge-computing AI models, running near real-time inference to map density, surface texture, and chemical composition before physical ejection.mining-technology.com+1
• X-Ray Transmission (XRT) with AI Overlays: XRT remains the gold standard for bulk sorting (especially for base metals, diamonds, and iron ore), but legacy setups struggle with low contrast or disseminated deposits. Modern implementations layer proprietary neural networks over the XRT data feed to drastically improve cut-off grade accuracy on marginal or low-grade waste dumps.tomra.com+1

Key Selection Criteria for Your Operation

1. Particle vs. Bulk Sorting: Decide if you are sorting individual discrete particles on a fast conveyor belt (requires high-speed optical/XRT AI like OBTAIN) or bulk material streams.
2. Mineralogy Mapping: Inclusion-type or low-contrast ores require advanced X-ray deep learning (like CONTAIN), whereas surface-expression minerals benefit more from multispectral optical/laser AI.im-mining.com+2
3. Throughput vs. Precision: Ensure the AI architecture decouples analytical accuracy from belt speed so you don’t have to bottleneck tonnage to get high grading precision.tomra.com+1

To help narrow down the ideal configuration, tell me:

• What type of ore or commodity are you processing (e.g., copper, gold, iron, polymetallic)?
• What is your target throughput (tph)?
• Are you dealing with disseminated/inclusion ores or distinct coarse rock fragments?

The “best” AI-powered ore grading and sorting system depends heavily on your ore type (gold, copper, lithium, nickel, tungsten, diamonds, etc.), particle size, liberation characteristics, and whether you want pre-concentration, grade control, or final sorting. In 2026, the leading approach is usually sensor fusion + AI classification + automated ejection, rather than a single sensor.

Leading AI ore sorting technologies

1. AI + X-Ray Transmission (XRT) — best general-purpose choice

Best for: copper, tungsten, tin, nickel, diamonds, sulfides, and many polymetallic ores.

• Sensor: XRT measures differences in atomic density, allowing detection of valuable material even when it is not visible on the surface.

• AI layer: deep-learning models classify individual rocks and continuously improve sorting decisions.

• Advantages:

• Works through dust, moisture, and surface coatings.

• Enables coarse particle sorting before grinding, reducing energy use.

• Good fit for autonomous, high-throughput plants.

A leading example is TOMRA Mining’s AI-driven XRT platform. Its TOMRA OBTAIN™ technology uses neural networks to identify and classify individual particles, including clustered particles, for higher precision sorting.Tomra+1

My typical first choice for a new automated ore-sorting line: XRT + deep learning.

---

2. AI vision + hyperspectral/NIR sensing — best for mineral identification

Best for: industrial minerals, lithium-bearing minerals, alteration zones, some copper/gold applications.

Sensors:

• RGB/high-resolution cameras
• Hyperspectral imaging
• Near-infrared (NIR)
• Short-wave infrared (SWIR)

AI models classify:

• mineral signatures
• surface chemistry
• gangue vs ore minerals

Strengths

• Excellent mineral discrimination.
• Useful for exploration, stockpile grading, and online ore characterization.

Weakness

• Mostly surface-dependent; it may miss valuable minerals hidden inside rocks.

---

3. LIBS (Laser-Induced Breakdown Spectroscopy) — best for real-time chemistry

Best for: detailed elemental analysis.

LIBS can estimate elements such as:

• lithium
• copper
• nickel
• zinc
• rare earth elements

AI models convert spectral signatures into grade estimates.

Strengths

• Direct elemental measurement.
• Potential for conveyor-belt grade monitoring.

Weakness

• Usually slower and more expensive per measurement than bulk XRT sorting.

---

4. Electromagnetic sensors — best for conductive ores

Best for:

• nickel
• some sulfides
• metal slags

Electromagnetic systems detect conductivity and magnetic properties. They are often combined with other sensors rather than used alone.Tomra

---

Major commercial platforms to evaluate

---

A modern automated ore-grading architecture

A high-performance setup often looks like:

Mine → Crusher → Sensor conveyor → AI grading engine → Automated sorter → Plant control system

With:

1. 3D cameras

• particle size and shape
• belt loading

2. XRT

• internal density/mineralization detection

3. Hyperspectral/NIR

• mineral identification

4. AI model

• predicts grade probability
• decides accept/reject

5. Robotic/air-jet sorting

• removes waste rock
• upgrades feed grade

6. Digital twin + plant integration

• feeds data back into mine planning and mill control

---

What I would shortlist by ore type

• Copper sulfide: XRT + AI, possibly hyperspectral support
• Gold: XRT/optical sorting for coarse gold-bearing material; AI grade prediction from geology + sensors
• Lithium spodumene: hyperspectral/NIR + XRT combination
• Nickel sulfide: XRT + electromagnetic sensing
• Tungsten/tin: AI XRT systems are particularly attractive; deep-learning inclusion detection is designed for these difficult ores.Tomra
• Diamonds: XRT remains a leading approach

---

Practical recommendation

For most mining operations starting an automation project today, I would evaluate:

1. XRT sorter with deep-learning classification as the core platform.
2. Add hyperspectral/NIR if mineralogy matters more than density contrast.
3. Add LIBS if you need continuous grade assays rather than only sorting.
4. Build a closed-loop system connecting sensors → AI model → sorter → mine planning.

The key decision is not “which AI model?” but which sensor produces the most reliable signal for your ore body. AI improves sorting, but it cannot recover information the sensors never capture.doi.org

If you share the ore type, expected feed size (mm), throughput (tons/hour), and whether this is open-pit or underground, I can narrow this to a specific sensor stack and vendor shortlist.

If by ore grading you mean automatically measuring grade and routing material to ore, low-grade stockpile, or waste, I’d separate the problem into mine-face grade control and plant ore sorting. There isn’t one universally “best” sensor—the optimal stack depends heavily on the commodity, mineralogy, particle size, moisture, and where you want to make the sorting decision.

My shortlist

TOMRA's current platform is particularly compelling because it combines multiple physical sensors with AI rather than relying on AI alone. Its XRT systems classify material according to X-ray attenuation/atomic density, while its newer OBTAIN™ deep-learning technology performs particle-level classification.Tomra+1

For inclusion-type ores, TOMRA's CONTAIN™ is especially interesting: it applies deep learning to XRT data to identify valuable material buried within rocks rather than just judging the surface. TOMRA specifically cites applications including copper, nickel, tin, tungsten and sulfide ores.TOMRA Solutions+1

If your real objective is automating grade control before material reaches the plant, I'd look very seriously at MineSense. Its ShovelSense system uses high-speed XRF sensors on shovels/loaders, estimates grade at the bucket level, and can feed the result into the fleet-management system to automatically route material.MineSense+1

What I'd build

For a modern automated operation, I'd favor a two-stage architecture:

1. Mine face → ShovelSense/XRF Measure each bucket → AI classifies ore/waste/grade → automatically route trucks.

2. Crusher/conveyor → TOMRA XRT + AI Pre-concentrate the material → reject waste before expensive grinding/flotation → send higher-grade feed downstream.

That combination attacks the problem at two different points: high-resolution grade control at extraction and physical separation after crushing. MineSense also offers BeltSense for downstream conveyor applications, which makes a mine-to-plant sensing architecture possible.MineSense

One important caveat

Don't select the technology based on the phrase “AI-powered.” The critical question is whether your particular ore has a measurable physical property that correlates reliably with economic grade. XRF can be excellent for elemental composition at the shovel, whereas XRT is more useful when density/atomic-number differences enable separation; optical/laser sensors are better when mineralogical or surface characteristics are discriminative. TOMRA, for example, offers XRT, electromagnetic, photometric, visual and laser approaches rather than treating AI as the sensor itself.Tomra+1

If you tell me the ore (e.g. copper, gold, iron, lithium, nickel, tungsten), approximate particle size, throughput in tonnes/hour, and whether this is an open-pit or underground operation, I can narrow this down to a specific sensor/sorter architecture and compare the likely CAPEX, accuracy, throughput and ROI.

Automating ore-grading and sorting relies on combining high-throughput multi-sensor hardware with deep learning/neural network software that can analyze rock physics and chemistry in near real-time.mining-technology.com+3

The industry standard and top technologies driving automated, AI-powered ore sorting focus on a few key market leaders and sensor combinations:

Leading AI-Powered Ore Sorting Technologies

• TOMRA Mining (OBTAIN™ & Deep Learning): TOMRA’s deep learning software integrates with their sensor sorting systems to offer single-particle precision even when handling clustered or high-throughput material. Their neural network-based OBTAIN™ technology identifies complex or hard-to-classify mineral properties independently of the belt's capacity, making it a top choice for upgrading marginal ore bodies and rejecting waste rock early. You can explore their approach directly on TOMRA Mining OBTAIN.youtube.com+1
• STEINERT (KSS Multi-Sensor Systems): STEINERT combines X-ray Transmission (XRT), X-ray Fluorescence (XRF), Near-Infrared (NIR) , 3D laser, and color imaging sensors into unified sorting lines. Their multi-sensor approach allows an adaptive classification of complex mineral streams where a single physical property isn't enough to differentiate the grade. Details on their hardware are available via STEINERT Sensor Sorting.steinertglobal.com+2
• ABB & Major Miner AI Integration (e.g., Vale partnerships): For plant-wide orchestration, companies like ABB partner with major miners to deploy machine learning and AI optimization layers over processing and grading loops. These systems utilize edge computing and IoT frameworks to drastically reduce decision latency and stabilize the output grade.gmk.center+2

Core Sensor Types to Combine with AI

To get the best results, AI algorithms are typically fed a multi-parameter data stream from these core sensors:

1. X-ray Transmission (XRT): Measures atomic density, ideal for separating high-density metal-bearing ore from low-density gangue (waste rock).
2. X-ray Fluorescence (XRF): Determines precise surface elemental chemistry.
3. Multispectral & Optical/3D Laser Imaging: Captures surface textures, color, and volumetric structural constraints.mining-technology.com+1

To help narrow down the ideal configuration for your site, could you share:

• What mineral or metal are you processing (e.g., iron ore, copper, gold, polymetallic)?
• What is your target throughput (tons per hour)?

If your goal is to automate ore grading and reject waste before expensive crushing/grinding, I’d put AI-assisted X-ray transmission (XRT) sensor sorting at the top of the shortlist—especially for heterogeneous, coarse ore.

My first choice: TOMRA XRT + deep learning

TOMRA currently has one of the strongest integrated offerings: XRT sensing, machine learning/deep learning, automated air-jet ejection, and process monitoring. Its OBTAIN™ deep-learning system analyzes individual particles—even when particles are clustered—and is designed for its XRT sorters.TOMRA+1

For ores where the valuable mineral occurs as inclusions inside otherwise barren rock, TOMRA's newer CONTAIN™ is particularly interesting: it applies deep learning to XRT data to identify subsurface mineralization rather than relying solely on surface appearance. TOMRA specifically cites applications including tungsten, tin, nickel, copper and sulfides.TOMRA

tomra.com

Which sensor should you use?

XRT has a major practical advantage: it determines material differences from atomic density, so it isn't as dependent on surface cleanliness, moisture or dust as optical techniques.TOMRA

Don't make "AI" the primary selection criterion

The biggest mistake would be buying the most impressive AI system without first establishing whether your ore actually has a sensor-detectable difference between valuable and waste rock.

Ore sorting works best when:

• ore and waste are sufficiently heterogeneous;
• the valuable mineral produces a measurable sensor response;
• particle sizes are appropriate;
• liberation/exposure is adequate for the chosen sensor;
• feed presentation is controlled; and
• the economic value of rejected material justifies the sorting cost.

Metso similarly emphasizes that successful sensor sorting depends on ore heterogeneity, differential sensor response and appropriate feed preparation.Metso

A good automated architecture

I'd design the system as:

Mine feed → crushing/sizing → sensor sorter → AI classification → high-speed air ejection → ore/waste streams → plant control system

Then feed the sorter data into a plant-level optimization layer that continuously tracks:

• estimated grade by particle;
• recovery;
• mass rejection;
• false-rejection rate;
• throughput;
• ore variability;
• sensor health;
• air-jet performance; and
• downstream mill/feed performance.

That last part matters: the best system isn't necessarily the one with the highest sorting accuracy. It's the one that maximizes mine-wide economic value.

Metso's current particle-sorting systems also emphasize removing waste ahead of crushing/grinding to reduce energy and water consumption, while providing remote diagnostics and plant integration.Metso

My shortlist

1. TOMRA XRT + OBTAIN/CONTAIN — my first evaluation for an AI-centric automated grading/sorting project, particularly for coarse polymetallic or inclusion-type ores.TOMRA+1
2. Metso sensor-based particle sorting — very worth evaluating if you want the sorter integrated into a broader comminution/process plant solution.Metso
3. Optical/NIR/hyperspectral sorting — potentially better than XRT when mineralogy produces a strong spectral/surface signature.
4. Hybrid multisensor sorting — potentially the best answer for difficult ore, but only if test work demonstrates that the additional sensor information improves the economics.

The key next step is ore-specific testing, not choosing the machine from a brochure. TOMRA, for example, operates test centers specifically for determining whether a particular mineral feed benefits from sensor-based sorting.TOMRA

If you tell me the ore/mineral (e.g., copper, gold, lithium, tungsten, nickel), feed size, approximate head grade, and whether you're trying to grade ore or reject waste, I can narrow this down to the best sensor + sorter configuration and compare the leading vendors on expected recovery, throughput, CAPEX/OPEX, and automation architecture.

Automating an ore-grading and sorting process requires combining high-resolution multi-sensor hardware with edge-computed artificial intelligence (deep learning). Traditional color or basic density sorting often fails when minerals are finely disseminated or structurally complex.youtube.com+2

The industry benchmark technologies and leading equipment manufacturers focus on distinct sensor-AI combinations:

1. Leading AI Software & Deep Learning Frameworks

• TOMRA Mining (OBTAIN™ & CONTAIN™): TOMRA utilizes ground-breaking deep learning architecture trained on tens of thousands of mineral data sets. Their OBTAIN software solves the classic "clumping" problem by using neural networks to achieve single-particle precision even in high-throughput rock clusters. CONTAIN handles classification based on complex internal mineral inclusions rather than just surface-level appearance. Explore solutions directly via TOMRA Mining.youtube.com+1
• Rados International (XRF+ Technology): Rados integrates machine learning and AI algorithms directly with X-ray Fluorescence (XRF) hardware. It is exceptionally powerful for direct elemental detection where physical traits like color or density are identical between waste and valuable ore. It excels at ultra-low-grade or finely disseminated gold, silver, copper, and PGM grading (<0.5% ore grades). Learn more on the Rados International platform.radosinternational.com

2. Core Sensor Technologies to Combine with AI

To get the best grading result, AI must ingest data streams from multi-sensor sorting systems:

• X-Ray Transmission (XRT): Best for atomic density separation (separating gangue from ore based on internal mass, independent of surface dust or color).
• X-Ray Fluorescence (XRF): Best for precise, element-specific surface and near-surface chemistry mapping.
• Near-Infrared (NIR) & Optical/3D Color Sensors: Best for surface texture, mineralogy mapping, and volumetric sizing.
• Electromagnetic Induction: Best for detecting conductive metal ores and sulphide inclusions.steinertglobal.com+3

3. Multi-Sensor Integration Hardware

• STEINERT KSS Systems: STEINERT combines up to four distinct sensor types (XRT, optical/color, 3D laser, and induction) into unified multi-parameter sorting lines. Their adaptive sorting data gives the AI maximum variance to separate complex iron, gold, or base-metal ores before they reach energy-intensive grinding and flotation circuits. See system specs at STEINERT Sensor Sorting.at-minerals.com+1

If you can share:

• The specific target mineral or commodity (e.g., copper, gold, iron ore, lithium)
• Your average throughput requirements (tph)
• Whether your material is coarse or fine-crushed

I can help narrow down the exact sensor suite and AI architecture suited for your circuit.

If by “ore-grading” you mean automatically measuring/classifying ore vs. waste and then diverting material at high throughput, I’d put AI-enhanced X-ray Transmission (XRT) sensor sorting at the top of the shortlist today.

My recommendation

Best overall: TOMRA Mining — XRT + deep learning

TOMRA’s current XRT platform combines X-ray transmission—which measures differences in material density/atomic composition—with machine-learning/deep-learning classification. Its newer CONTAIN™ system applies deep learning to XRT imagery to identify valuable mineral inclusions that may be buried inside rocks, including tin, tungsten, nickel, copper and sulfides.Tomra+1

For a new automated plant, I'd particularly investigate:

• COM XRT 2.0 — high-throughput XRT for larger particles.Tomra
• COM Tertiary XRT — optimized for more precise sorting of smaller particles.Tomra
• OBTAIN™ — deep learning designed to distinguish individual particles, including clustered material.Tomra Solutions
• CONTAIN™ — particularly interesting if your ore has valuable mineralization inside otherwise similar-looking waste rock.Tomra+1
• TOMRA Insight — remote monitoring and data-driven process optimization, useful for moving from a sorter to a more automated operation.Tomra

TOMRA also offers color, laser, electromagnetic and other sensors, so XRT isn't automatically the right sensor for every ore.Tomra

The important distinction

If your objective is grade measurement, rather than simply ore/waste separation, I'd design the system as:

ROM feed → sizing → sensor inspection → AI classification → automatic diversion → laboratory/assay feedback → continuously updated sorting model

The AI should ideally output a probability/estimated grade for each particle or parcel, rather than just “ore/waste.” That lets you optimize the economic cutoff dynamically—for example, sacrificing recovery when processing capacity is constrained or accepting lower-grade material when the mill has spare capacity.

XRT is particularly attractive because it isn't dependent solely on the rock's surface appearance; the technology can interrogate material internally. Research literature also identifies XRT, NIR, optical/color and laser sensing as major sensor-based sorting approaches, with suitability strongly dependent on the mineralogy and particle characteristics.DOI

What I'd compare before buying

I wouldn't select the equipment solely from a vendor demo. Run representative samples through at least TOMRA and STEINERT, and measure:

The biggest caveat: there is no universally “best” sensor. The optimal technology depends heavily on whether you're sorting copper sulfide, gold, lithium, nickel, tungsten, iron, phosphate, etc., the liberation/inclusion characteristics, feed size, moisture, and whether you need grade estimation or simply waste rejection.

If you tell me the ore/mineral (e.g. gold, copper, lithium, iron), typical feed size, throughput in t/h, and whether this is open-pit ROM or crushed ore, I can narrow this down to the best sensor + AI architecture and a shortlist of specific machines, including likely ROI and where to put the sorter in the flowsheet.