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

Yes. If your goal is to automate ore grading and reject waste before it reaches the main processing circuit, the best technology is usually not a generic “AI camera.” It is a sensor-based ore sorter with AI/deep-learning classification, selected around the mineralogy and particle size of your ore.

My current shortlist

The technology I'd investigate first: XRT + deep learning

For a serious automated ore-grading project, X-ray transmission (XRT) is particularly attractive because it isn't dependent solely on what is visible on the rock surface. XRT measures X-ray attenuation related to material density/atomic composition, and can operate despite surface dust, dirt and moisture.

TOMRA has now added two interesting AI layers:

• OBTAIN™ — deep learning that classifies individual particles, including particles in clustered material, allowing the sorter to make more precise ejection decisions.
• CONTAIN™ — deep learning applied to XRT data to detect valuable mineral inclusions inside rocks. TOMRA specifically identifies applications including tin, tungsten, nickel, copper and sulfides.

That's a meaningful distinction: you're moving from “does this rock look like ore?” toward “what does this individual rock contain, and should I send it to the mill?”

There is also real-world deployment evidence: TOMRA reports current projects combining XRT with OBTAIN and CONTAIN at Soma Gold, while its customer portfolio includes XRT applications for tungsten, tin, chrome, lithium and other ores.

But don't choose the sensor before testing your ore

The ore/mineralogy is the decisive variable. I'd structure the project as:

Mine feed → crushing/screening → sensor scanning → AI classification → pneumatic ejection → graded ore + waste

Then build the AI model around assay-confirmed samples, rather than trying to have an AI model infer “grade” directly from arbitrary imagery.

For example:

• Dense vs. low-density ore: XRT
• Conductive mineral: electromagnetic/induction
• Surface mineral/color/texture: laser/optical
• Mineralogical composition: NIR/hyperspectral
• Complex inclusions: XRT + deep learning
• Maximum robustness: potentially sensor fusion rather than one sensor

TOMRA itself offers multiple sensing modalities, including XRT, electromagnetic and photometric/laser sensing.

What I'd do before buying anything

Run a representative bulk-ore test with the vendors. TOMRA, for example, has dedicated mining test centers specifically for determining whether sensor sorting will work on a particular material.

Measure:

1. Feed size distribution
2. Head grade and grade variability
3. Mineral liberation size
4. Ore/waste density contrast
5. Moisture and surface contamination
6. Target recovery %
7. Waste rejection %
8. Final concentrate grade
9. Throughput tonnes/hour
10. $/tonne operating and capital cost

The key economic metric isn't simply sorting accuracy. It's whether pre-concentration lets you reject enough barren material early enough to reduce milling, grinding, reagent, water and tailings costs while retaining economically valuable metal.

If you tell me the ore you're grading (e.g. gold, copper, lithium, iron, tungsten, nickel), particle-size range, approximate throughput in t/h, and whether you're trying to grade ore or separate ore from waste, I can narrow this to the best sensor architecture and specific commercial machines.

Here are top web results for exploring this topic:

Mining Technology·https://www.mining-technology.com**AI**-**powered ore sorting** - Mining Technology An ore sorting technique that uses an innovative combination of AI analytics of sensor (laser and optical) data and advanced AI machine vision analysis of (X-ray, 3D and multispectral) images has emer TOMRA·https://www.tomra.com Mining - sensor -based sorting - TOMRA Maximize efficiency and reduce costs with TOMRA's sensor-based ore sorting technology. Enhance output, quality, and sustainability in mining operations. YouTube·https://www.youtube.com How TOMRA Mining Uses AI & Deep Learning to Revolutionize Ore ...Artificial Intelligence is changing the mining industry. Explore how TOMRA Mining integrates AI and deep learning into its sensor-based sorting solutions. Take a closer look at how integration of the International Mining·https://im-mining.com TOMRA Mining takes next AI-powered ore sorting step with CONTAIN TOMRA Mining takes next AI-powered ore sorting step with CONTAIN. Posted on 13 Jun 2025. TOMRA Mining has launched CONTAIN™, a solution that, it says, represents the next breakthrough in AI-powered or ScienceDirect.com·https://www.sciencedirect.com**Artificial intelligence** of mineral processing process : A review of ...With the advancement of artificial intelligence technology, intelligent mineral processing has become an inevitable trend in the development of the industry. In this article, various models and their Imubit·https://imubit.com 5 AI -Driven Methods to Improve Critical Mineral Separation Efficiency 1. Predict Optimal Process Conditions Before They're Needed. Predictive AI learns from your plant data and live sensor feeds to spot subtle ore-quality shifts minutes—or even hours—before they upset r Dataintelo·https://dataintelo.com**Automated Ore Sorting** Systems with Robotics Market - Dataintelo Another key driver is the technological evolution in sensor-based sorting and robotics, which has made ore sorting more accurate, reliable, and scalable. The incorporation of sophisticated sensors suc steinert·https://steinertglobal.com Efficiently optimise ore sorting - steinert ... process at an early stage to save energy and resources and to unlock more enriched ore more effectively. With preconcentration there are customers who can double the grade of their material. The S YouTube·https://www.youtube.com TOMRA Mining Presents: OBTAIN™ - AI-powered ore sorting OBTAIN™ - A groundbreaking deep learning technology for TOMRA's sorters, revolutionizes ore sorting by enabling precise detection and classification of individual particles, even when clustered. OBTAI AZoMining·https://www.azomining.com Improving Sustainability and Efficiency in Ore Sorting with Sensors This process allows mining companies to extract the most value by prioritizing high-grade ore for energy-intensive downstream processing while diverting gangue to waste piles early on. Traditionally,

If by “ore-grading” you mean automatically determining grade and then routing ore/waste without relying on manual sampling, I’d separate the problem into two layers:

1. Grade measurement / characterization — determine what each bucket, truckload, or rock contains.
2. Sensor-based sorting — physically separate ore from waste.

There isn't one universally best sensor. The right choice depends heavily on the commodity, mineralogy, particle size, moisture, throughput, and whether you want bucket-level grade control or rock-by-rock sorting.

My shortlist

For an operating mine, I'd look very closely at these systems

1. TOMRA Mining — my leading candidate for automated rock sorting

TOMRA currently has one of the more sophisticated combinations of sensor sorting and AI. Its XRT systems can classify material based on density/atomic characteristics, while its OBTAIN™ deep-learning system performs single-particle classification, including when particles are clustered.

More interestingly, CONTAIN™ applies deep learning to XRT imagery to identify valuable inclusions hidden within rocks. TOMRA specifically cites applications including tin, tungsten, nickel, copper and sulfides.

That makes the combination:

XRT → deep-learning classification → high-speed ejection

a particularly compelling architecture for automated pre-concentration.

2. STEINERT KSS — my leading alternative for multi-sensor sorting

STEINERT's KSS platform can combine XRT, color, 3D/laser and inductive sensing. That is attractive if your ore cannot be reliably discriminated using one physical property.

The advantage is essentially sensor fusion: XRT sees through the rock, optical/3D sensors characterize its surface and geometry, and induction adds information about conductive material.

3. MineSense — if “grading” happens at the shovel rather than on a sorting belt

This is a different problem and potentially a better fit if you're trying to automate ore/waste classification during excavation. MineSense's ShovelSense uses high-speed XRF sensors mounted on mining shovels/loaders and uses machine learning to estimate grade during each digging cycle. The resulting classification can be sent to fleet-management systems to route the load.

Its BeltSense system extends that concept onto conveyors and can feed grade information into plant control systems.

What I'd build

For a new automated system, I'd favor a sensor-fusion architecture rather than betting everything on an AI camera:

Mine face → XRF grade sensing
→ AI ore/waste classification
→ autonomous truck routing

Primary crusher → screening into appropriate size fractions

Sorting conveyor → XRT + optical/3D + possibly NIR/induction
→ deep-learning classification
→ pneumatic ejection

After sorting → continuous XRF/BeltSense-type measurement
→ mill-feed grade prediction
→ automatic blending/process control

This is important because AI is the decision layer, not the fundamental measurement. You want physically meaningful sensor data feeding the model. Recent research is also moving toward combining XRF spectra with deep-learning architectures rather than treating XRF as a standalone classifier.

If you tell me your ore, I can narrow this down dramatically

Give me these five things:

• Commodity: copper, gold, iron, lithium, nickel, tungsten, etc.
• Ore mineral / host rock
• Typical feed size: e.g. 20–100 mm
• Throughput: tonnes/hour
• Whether you want bucket/truck grading, conveyor sorting, or both

With those, I can recommend the specific sensor combination and likely vendors, and lay out a practical automated ore-grading/sorting architecture with estimated CAPEX/OPEX considerations.

Automating an ore-grading and sensor-based sorting (SBOS) process relies heavily on combining high-resolution physical sensors with edge-computed deep learning algorithms . Rather than a single "best" piece of equipment, the industry standard relies on integrated ecosystems tailored to the specific mineralogy, throughput, and whether you are sorting bulk material at the extraction face or pre-concentrating at the mill.

The leading AI-powered sensor and sorting technologies dominating the market include:

• TOMRA Mining (OBTAIN™ & CONTAIN™ with XRT) : TOMRA leads the deployment of industrial deep learning for mineral sorting . Their OBTAIN™ software uses convolutional neural networks for single-particle precision—even when rocks are clustered on high-speed conveyor belts—dramatically increasing product purity without sacrificing throughput. Their CONTAIN™ platform is purpose-built for complex, inclusion-type ores (such as tungsten, nickel, or tin) by analyzing real-time X-Ray Transmission (XRT) imagery to identify internal mineral features invisible to standard sensors. Learn more about their approach via TOMRA Mining AI Solutions.
• STEINERT (XSS T EVO 5.0 & KSS EVO 6.0 Series) : STEINERT specializes in multi-sensor sorting platforms combining dual-energy X-ray Transmission (XRT), X-ray Fluorescence (XRF), Near-Infrared (NIR), and 3D optical/laser systems. Their hardware excels at early-stage pre-concentration—stripping waste rock and low-concentration ore out of the stream early to drastically lower downstream water, energy, and reagent costs. Explore their lineup at STEINERT Ore Sorting.
• MineSense & Plotlogic (Pit-to-Mill Grade Intelligence) : If you need grading and sorting intelligence before material even hits the processing plant or conveyor, companies like MineSense and Plotlogic (with its OreSense platform) integrate hyperspectral imaging, laser sensors, and real-time AI modeling right at the extraction face and on shovels. This shifts the automation upstream, ensuring accurate grade control during initial excavation.

Core Sensor Technologies to Consider

• X-Ray Transmission (XRT) : Best for sorting based on atomic density; ideal for bulk pre-concentration of base metals, precious metals, and diamonds regardless of whether the mineral is trapped inside the rock or on the surface.
• X-Ray Fluorescence (XRF) : Best for surface elemental composition analysis.
• Near-Infrared (NIR) & Optical/Laser : Best for color grading, crystalline structures, and non-sulfide surface variations.

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

• What mineral or commodity are you processing?
• Are you looking to sort underground/at the face , on a conveyor pre-concentration belt , or at the mill feed?
• What is your target feed size range (mm) and throughput (tph)?

If your goal is to automate ore grading and reject waste before it reaches the mill, I’d focus less on a generic “AI sensor” and more on a sensor-based ore-sorting system with AI/deep-learning classification.

My leading choice: TOMRA XRT + AI

For a modern automated operation, TOMRA’s X-Ray Transmission (XRT) platform combined with its OBTAIN™ and, where applicable, CONTAIN™ deep-learning software is one of the strongest options I’d evaluate first.

• XRT looks through the rock rather than relying only on its surface, making it useful for density/atomic-composition differences and for mineralization hidden inside particles.
• OBTAIN™ uses deep learning to identify individual particles, including clustered material, and make higher-precision sorting decisions at production throughput.
• CONTAIN™, introduced for commercial applications in 2025, applies deep learning to XRT imagery to identify subsurface inclusions—particularly relevant to tungsten, tin, nickel, copper and sulfide ores.
• TOMRA also has NIR, visual, electromagnetic and photometric sensing, so the system can be tailored rather than relying on XRT alone.

For example, TOMRA reports XRT applications for sulfide ores in the roughly 8–80 mm range and says its systems can achieve substantial feed-grade upgrades, although the actual result is highly deposit-specific.

The main alternative I'd benchmark: STEINERT

STEINERT is particularly interesting if you want a multi-sensor architecture rather than primarily XRT + AI.

Its KSS platform can combine:

XRT + color + 3D/laser + induction, with other configurations incorporating NIR or XRF.

That's attractive for heterogeneous ore where no single physical property cleanly distinguishes ore from waste. STEINERT also has an “Intelligent Declustering” algorithm that uses image processing to virtually separate touching particles before making sorting decisions.

What I'd choose by ore type

STEINERT's technology portfolio explicitly supports XRT, XRF, NIR, induction, 3D, laser and color sensing, while TOMRA offers XRT, VIS, color, EM and other sensors.

One important distinction: grading vs sorting

If by “ore grading” you mean continuous measurement of the actual metal grade of every rock, don't assume an ore sorter is equivalent to an assay lab.

I'd architect the system as:

Mine → crusher → screen → sensor scan → AI classification → pneumatic ejection → ore/waste streams → automated assay/QA feedback

The AI should ultimately optimize grade–recovery, not simply classify rocks as “ore” or “waste.” TOMRA's CONTAIN, for example, allows operators to adjust the grade/recovery threshold based on the classification probability.

What I'd do before buying anything

Don't select the machine from a brochure. Send representative ore samples from your deposit to both vendors for production-scale sorting trials. Sensor performance depends heavily on mineralogy, liberation, particle size, moisture, surface coating and the contrast between ore and waste. TOMRA itself emphasizes ore testing because the appropriate sensor configuration is application-specific.

I'd run a head-to-head trial measuring:

1. Grade upgrade
2. Recovery of valuable metal
3. % mass rejected
4. False-rejection rate
5. Throughput t/h
6. Particle-size operating window
7. Compressed-air and electrical consumption
8. Performance as ore grade/mineralogy changes
9. AI model stability after commissioning
10. $/tonne and payback

Bottom line: if you have a sulfide, tungsten, tin, nickel or similar deposit where mineralized rock has a meaningful X-ray/density signature, I'd put TOMRA XRT + OBTAIN/CONTAIN at the top of the shortlist, with STEINERT's multi-sensor KSS as the key competing system to benchmark. The actual “best” system should be determined by a controlled sample trial, not the vendor's headline AI capability.

If you tell me the ore (e.g., gold, copper, lithium, tungsten), typical head grade, particle-size range, and target throughput in tonnes/hour, I can narrow this to a specific sensor architecture and sorter configuration.

Automating an ore-grading and pre-concentration process relies on matching the right sensor physics with advanced artificial intelligence (specifically deep learning and computer vision). Rather than relying on a single "best" technology, the industry standard centers on multimodal sensor platforms powered by neural networks capable of analyzing thousands of rock fragments per second on high-speed conveyor belts.

Leading technologies and system architectures divide into specific categories:

• TOMRA Mining (OBTAIN™ & CONTAIN™ Ecosystem) : Widely recognized as a market leader in sensor-based sorting. Their ecosystem uses convolutional neural networks for real-time analysis. OBTAIN™ handles single-particle precision by separating clustered rocks on the belt without sacrificing throughput, while CONTAIN™ uses deep learning to classify complex inclusion-type ores (such as tin, nickel, and tungsten) based on subsurface mineral signatures captured via X-ray imagery.
• STEINERT (KSS Multi-Sensor Concept) : A powerhouse in flexible modular sorting. The STEINERT KSS system combines X-ray Transmission (XRT) , 3D laser scanners, color optical cameras, and electromagnetic induction sensors into a single unified framework. Their multi-sensor approach allows the AI to cross-reference density, atomic weight, and surface features simultaneously.
• Hyperspectral & Laser Systems (e.g., Hypermine) : Emerging technologies that combine laser illumination and hyperspectral sensing with machine learning algorithms. These systems excel at early-stage stockpile management, identifying chemical contaminants (like alumina or sulfates) and optimizing shovel-by-shovel or truckload routing before the ore ever touches crushing or grinding circuits.
• PGNAA (Prompt Gamma Neutron Activation Analysis) : Ideal for bulk elemental analysis (such as cement, coal, and major metal grades) on a continuous mass-flow basis rather than discrete particle-by-particle tracking. When integrated with predictive machine learning models, it stabilizes overall plant feed consistency.

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

• What type of ore or mineral you are processing (e.g., copper, gold, base metals, industrial minerals)?
• Are you looking for particle-by-particle sorting (conveyor belt sorting) or bulk stream analysis?
• What is your target throughput (tons per hour)?

If by “ore grading” you mean automatically determining whether each rock/particle is valuable ore, its approximate grade, and then routing it to different processing streams, I’d build the system around sensor-based sorting + machine learning, rather than trying to replace laboratory assays with a camera-only AI model.

My first choice: AI + XRT

For many hard-rock operations, X-ray Transmission (XRT) combined with deep learning is currently the strongest starting point.

• XRT sees through the particle and measures density/atomic-density contrasts, so it can detect valuable material that isn't exposed on the surface.
• AI can then classify the X-ray images and make particle-level decisions.
• TOMRA Mining currently has particularly interesting AI capabilities: CONTAIN™ uses deep learning on XRT imagery to identify valuable inclusions buried inside rocks, including applications involving copper, nickel, tungsten, tin and sulfides.
• Its OBTAIN™ system uses deep learning to identify individual particles even when particles are clustered, allowing the sorter to optimize purity versus throughput.

A strong alternative is STEINERT. Its newer systems can combine XRT + color + 3D/laser + inductive sensing, rather than relying on one sensor.

Which sensor should you use?

The important distinction is that “best AI sensor” depends heavily on the ore body. XRT isn't automatically best for every deposit.

What I'd automate

I'd design the plant as:

Crush → screen → sensor scan → AI classification → high-speed air ejectors → ore / middling / waste → feedback from assays

The AI shouldn't merely answer “ore or waste?” Ideally it produces something like:

Particle #18472 → Cu-bearing → estimated grade class 1.2–1.8% → 87% confidence → ore stream

Then laboratory assays are continuously fed back into the model so that the sorter learns the relationship between sensor signatures and actual grade.

For a copper, nickel, lithium, gold, tungsten, or polymetallic operation, I would run a representative bulk sample through several sensor technologies before buying equipment. The decisive test isn't the vendor's headline accuracy—it's recovery vs. rejection at your actual particle-size distribution and throughput, followed by the resulting economics.

If you tell me the ore/mineral (e.g. copper, gold, lithium), typical particle size, throughput in tonnes/hour, and whether this is underground or open-pit, I can narrow this down to the 2–3 best sensor/sorter systems and a proposed automated grading architecture.

Automating an ore-grading and pre-concentration process relies on matching the right sensor physics with advanced artificial intelligence (AI) and deep learning algorithms . Rather than looking for a single "best" device, the industry standard relies on integrated ecosystems where multi-sensor hardware feeds real-time data into neural networks.

Leading AI-Powered Ore Sorting Technologies

• TOMRA Mining (OBTAIN™ and CONTAIN™ Ecosystem): Widely recognized as a market leader in sensor-based sorting, TOMRA integrates deep learning architectures (convolutional neural networks) directly into its sorting lines.
  • OBTAIN™ uses neural networks for single-particle precision, breaking down clustered rock flows so individual pieces are classified and ejected accurately even at high throughputs.
  • CONTAIN™ is purpose-built for complex "inclusion-type" ores (like nickel, tin, and tungsten), using real-time X-ray imagery analysis to detect subsurface mineral profiles that traditional algorithms miss. Operators can dynamically adjust grade-recovery thresholds via software controls. More details can be found directly via TOMRA Mining.
• Steinert & Redwave (XRF and XRT Systems): These providers specialize in X-ray Fluorescence (XRF) and X-ray Transmission (XRT) sensors backed by high-speed processing. XRF provides elemental surface composition data in milliseconds, which machine learning models use to predict internal grade compliance and direct high-value fractions away from waste streams.
• Ganzhou HPY Technology (Dual-Energy X-Ray): Widely implemented in large-scale base metal operations (such as lead-zinc and copper circuits), their dual-energy X-ray equipment combined with smart analytics effectively rejects massive tonnages of barren rock prior to wet plant processing.

Key Sensor Technologies to Consider

1. X-Ray Transmission (XRT): Measures the atomic density of particles. It is the gold standard for bulk separation of waste from valuable ore across high tonnages.
2. X-Ray Fluorescence (XRF): Ideal for surface-level direct elemental identification (e.g., copper, nickel, iron). When paired with machine learning, it correlates surface signatures to volumetric grade.
3. Near-Infrared (NIR) & Laser/Optical: Excellent for identifying specific mineralogical surface textures, color, or structural matrices (like sorting quartz from waste rock).
4. PGNAA (Pulsed Fast-Thermal Neutron Activation): Used for bulk stream elemental analysis where whole-belt chemistry is monitored continuously.

To help narrow down the optimal hardware and AI configuration for your site, could you share:

• What mineral or commodity are you processing?
• What is the average particle size of the feed?
• Are you aiming primarily for waste rejection (pre-concentration) or precise grade control?

Automating ore-grading and sorting requires pairing high-speed sensor hardware with modern AI and deep learning architectures that can process petabytes of mineralogical data in real-time.

The industry benchmark relies on sensor fusion (combining multiple physical/chemical data streams) driven by advanced neural networks.

Leading AI-Powered Ore Sorting Technologies

• TOMRA Mining (OBTAIN™ and CONTAIN™ Ecosystem): TOMRA stands out as an industry leader utilizing deep learning via convolutional neural networks. Their OBTAIN™ software uses neural networks to achieve single-particle precision even when rock particles are clustered on high-throughput belts. Their newer CONTAIN™ platform is purpose-built for complex "inclusion-type" ores (detecting subsurface or hidden mineralizations of tungsten, tin, nickel, and copper that traditional optical or basic XRT systems miss).
• Metso (Sensor-Based Particle Sorting): Offers robust modular sorting solutions designed to integrate into primary crushing and grinding circuits. Their multi-sensor integration targets broad spectrums—from color/optical and IR to high-capacity atomic density detection—drastically reducing downstream energy and water waste.
• Steinert (KSS Cliff Image Recognition / XRT): Known for robust sorting machinery (like the KSS series) combining 3D laser, induction, color, and X-ray sensors. Their smart sorting algorithms reduce the need for highly selective manual or basic mechanical mining by cleanly rejecting barren gangue early.
• Ganzhou HPY Technology (Dual-Energy X-Ray): Widely utilized in massive base-metal and polymetallic applications (such as lead-zinc and copper mines) to effectively pre-concentrate low-grade feed.

Core Sensor Technologies to Combine with AI

1. X-Ray Transmission (XRT): The gold standard for sorting based on effective atomic density. When backed by AI (like TOMRA's deep learning layers), XRT can peer through the rock matrix to evaluate internal mineral content rather than just surface composition.
2. Near-Infrared (NIR) & Hyperspectral Imaging: Excellent for surface mineralogy, identifying specific molecular bonds and mineral phases rapidly.
3. Laser and Optical Sensors: High-resolution multi-spectral cameras capture surface color, texture, and structural features, which deep learning models use to classify rock types instantly.
4. Laser-Induced Breakdown Spectroscopy (LIBS) / X-Ray Fluorescence (XRF): Used for direct elemental assaying on a particle-by-particle basis, providing exact chemical fingerprints.

To help narrow down which specific architecture fits your site, could you share:

• What is the target mineral or commodity you are grading/sorting?
• What is your estimated throughput (tph) and average particle size?
• Are you looking to upgrade crusher feed (pre-concentration) or process tailings/low-grade stockpiles?

The best solution depends much more on your ore type and mining method than on the AI software itself. Today's leading systems don't rely on a single "AI sensor"—they combine multiple sensors with machine learning to classify each rock fragment in milliseconds and then eject waste using compressed air or mechanical diverters.

Here's how I'd evaluate the technologies:

Leading commercial suppliers

Several companies are considered leaders in sensor-based ore sorting:

• TOMRA — Generally regarded as the industry benchmark for XRT-based sorting with AI-assisted classification, especially for hard-rock mining.
• STEINERT — Strong portfolio combining XRT, magnetic, induction, and optical sensors.
• MineSense — Focuses on real-time bulk ore sensing at the shovel and conveyor rather than particle sorting.
• NextOre — Specializes in bulk sensing for copper and other conductive ores.
• Comex Group — Offers AI-enabled multi-sensor systems including hyperspectral and XRT.
• HPY — Multi-sensor AI sorting with XRT and optical imaging.

If your goal is automated ore grading

Many operations now use a layered approach:

1. Face or shovel sensing (bulk grade estimation)
2. Conveyor belt analyzers
3. Particle ore sorting after primary crushing
4. Continuous AI model updates from assay results

This gives much better overall control than relying only on a sorter after crushing.

My recommendations by commodity

• Copper: XRF or XRT, often combined with bulk sensing (MineSense or NextOre).
• Gold: XRT if there is sufficient density contrast; otherwise combine optical and hyperspectral methods.
• Iron ore: Hyperspectral plus magnetic sensing can work well.
• Lithium (pegmatites): Hyperspectral plus XRT is increasingly common.
• Tungsten: XRT remains one of the strongest options.
• Industrial minerals: Hyperspectral imaging is often the preferred sensor.

What usually delivers the biggest ROI

The AI itself is rarely the deciding factor. The biggest gains typically come from:

• Rejecting waste before grinding (reducing energy and milling costs)
• Increasing mill feed grade
• Lowering water and reagent consumption
• Extending mine life by making lower-grade material economic

Operations see the best economics when the ore has clear physical or mineralogical differences from waste and when comminution is a major operating cost.

If you can tell me:

• the commodity (gold, copper, lithium, iron, etc.),
• whether it's open pit or underground,
• your approximate throughput (tph), and
• the particle size after primary crushing,

I can recommend a specific sensor suite and vendors that are the best fit for your operation.