Key Takeaways
- AI ore sorting builds on sensor-based sorting, a mature, high-adoption category. TOMRA and Steinert together hold over 40% of the market.
- The AI sits in the real-time classification logic that decides, particle by particle, whether a rock is ore or waste. This has moved well beyond simple threshold rules.
- TOMRA’s AI-powered advancements (OBTAIN, CONTAIN) target throughput (doubling sorting capacity) and fine inclusion detection for base metal sulphides, including copper.
- Sorting is a bulk pre-concentration step. Rejecting waste before the mill cuts downstream energy, water and tailings volume, so it is more than a grade-control tool.
TL;DR
AI ore sorting means installing a sensor-based sorter from TOMRA or Steinert KSS ahead of your mill. Classification software then decides in real time which rocks to keep and which to reject. Start with a test-center trial on your own ore, because sortability varies widely between deposits.
What Is AI Ore Sorting?
Ore sorting separates individual rocks into ore and waste before they reach the grinding circuit. Crushed rock passes a sensor on a belt or chute. The sorter then fires short bursts of compressed air to push waste particles out of the stream. What remains is a smaller, higher-grade feed for the plant.
Older sorters used a simple rule, such as a density or brightness threshold. AI ore sorting replaces that rule with a trained classification model. The model learns from labeled particles what ore and waste look like to the sensor. As a result, it handles ambiguous cases much better, such as mixed particles or ore that differs only subtly from waste.
How Do I Automate Ore Sorting With AI?
The work breaks into three parts: the sensor, the classification model and the business case.
Choosing the sensor for AI ore sorting
Most sorters in mining pass crushed material past an X-ray transmission (XRT) sensor array. XRT images the internal composition of each particle. Other sensors read surface colour, conductivity or near-infrared response, and some platforms combine several. The right choice depends on what physically separates your ore from waste. Our guide to ore sorting sensors covers that decision in detail.
Training the classification model
The classification layer is where the AI lives. Modern systems use learned models trained on labeled particle data. These handle fine-grained inclusions, mixed particles and ore types with subtle signal differences from waste rock.
TOMRA has pushed specifically into AI-powered capability with two named features. OBTAIN can roughly double sorting capacity by speeding up the classification decision. CONTAIN targets inclusion-type ores, explicitly including copper and other base-metal sulphides, where the valuable mineral occurs as fine inclusions. This matters for copper operations, because copper sulphide mineralization is often this hard-to-sort, fine-inclusion case.
Building the business case
The business case is usually framed as pre-concentration. Rejecting barren rock before the mill cuts the tonnage that needs grinding, along with the energy and water consumed. It also cuts the tailings volume produced. That is a different value proposition from grade-control sensing further down the line, such as a conveyor belt analyzer. A typical adoption path starts with a test-center trial. TOMRA runs dedicated test centers for this, using representative ore samples from your deposit.
What to Expect From AI Ore Sorting
Sorting does not suit every deposit. It works best when ore and waste occur as separate particles after crushing. By contrast, finely disseminated ore spread evenly through the rock gives the sensor little to separate. Therefore, the test-center trial is the real decision point, not the brochure.
Sorters also work on coarse crushed rock, so they sit after crushing and before grinding. Fine material usually bypasses the sorter. Plan for extra crushing and screening capacity, plus a waste stream that needs its own handling. Finally, expect some valuable metal to report to the reject stream. The trial should quantify that loss against the energy and water savings.
FAQs
How does AI ore sorting work?
A sensor scans each crushed rock as it passes on a belt or chute. A trained classification model decides whether the rock is ore or waste. An air jet then ejects the waste, all within a fraction of a second.
Which companies make AI ore sorting systems?
TOMRA and Steinert are the two dominant vendors, holding over 40% of the sensor-sorting market together. TOMRA offers AI features such as OBTAIN and CONTAIN. Steinert’s KSS platform supports combined sensors for more complex ores.
Does AI ore sorting work for copper?
It can. Copper sulphides often occur as fine inclusions, which are hard for simple threshold rules to detect. TOMRA’s CONTAIN feature targets exactly this case. However, you still need a trial on your own ore to confirm sortability.
What are the benefits of AI ore sorting?
The main benefit is pre-concentration: less waste rock reaches the mill. That reduces grinding energy, water use and tailings volume. It can also raise the head grade the plant receives.
Try It With Geocluster
Whether ore sorting makes sense for your deposit depends on ore type, mineralization style and grade distribution. Geocluster is designed to help you work through that deposit-specific evaluation before committing to a vendor trial. You can also read how the Geocluster Research Harness works.