Deep learning technology designed to enhance the recovery of inclusion-type ores that are difficult to detect using traditional sorting methods is the latest evolution in AI-driven sorting solutions.

TOMRA Mining recently unveiled its proprietary CONTAIN solution that uses convolutional neural networks to perform real-time analysis of X-ray imagery, visually classifying rocks based on the probability of subsurface ore mineral inclusions. These include complex mineralisations such as in tungsten, nickel and tin ores – materials that traditionally result in high misclassification or excessive product loss.

 “With CONTAIN, operators can dynamically adjust the grade-recovery threshold via a touchscreen interface, enabling precise control over yield and product specifications,” says  TOMRA Mining software team lead Stefan Jürgensen.

By analysing the structure of each rock using advanced deep learning algorithms, the system identifies subtle mineralogical patterns that indicate the presence of valuable metals such as tungsten, nickel or tin. Each rock is assigned a probability score based on its likelihood of containing mineralisation below the surface, enabling precise, data-driven sorting decisions. This capability allows mining operations to adapt their strategies in real time – whether the goal is to maximize concentrate grade, minimise valuable material loss, or align with processing cost constraints.

Built for industrial-scale performance, the system does not rely on specific throughput or spacing on the belt, ensuring that accuracy is maintained – even in dense, fast-paced input streams. It is especially effective in high-volume processing plants where consistency, speed and recovery rates are critical to profitability.

CONTAIN has been engineered to handle a spectrum of ore grades – from high-value deposits to low-grade, inclusion-rich rocks that have historically been difficult to process efficiently.

“Existing technologies can be configured to detect low-grade material in such ores, but this results in a high quantity of waste rocks being sorted into the product stream, diluting the product beyond economic viability. CONTAIN is exceptionally accurate in evaluating the value of a rock, making sorting thresholds for such relatively low-grade ores economically viable,” explains Jürgensen.

The system was designed to complement TOMRA’s sensor-based sorting ecosystem, working in concert with the COM XRT and OBTAIN systems to deliver a comprehensive, multi-layered approach to ore processing.

This integration is designed to give mining operations the ability to fine-tune performance across the entire sorting line, with data-driven control and real-time responsiveness. It also simplifies system scalability, allowing plants to evolve their capabilities without overhauling existing infrastructure.