Achieving high purity aluminium circularity through precision AI sorting

The global aluminium industry is evolving rapidly as manufacturers increasingly commit to ambitious net zero targets. This shift has driven a demand for high quality secondary aluminium that has outpaced the industry's ability to provide it using traditional recovery methods. Today, the challenge is no longer just about volume but about purity. To replace primary aluminium in high end applications, recyclers must achieve alloy specific separation while managing increasingly complex scrap streams. Success depends on the ability to effectively generate these precise scrap qualities, making innovative sorting technologies a pre-requisite for high quality material recovery and long-term sustainability.

From atomic density to chemical precision

Artificial intelligence (AI) has become the key enabler in transforming the capabilities of sensor-based sorting. For decades, x-ray transmission (XRT) has been the industry standard for separating heavy metals from aluminium by measuring atomic density. TOMRA’s industry-leading X-Tract, for example, sorts shredded mixed non-ferrous metals (Zorba) to produce high-purity aluminium scrap (Twitch), while also enabling the sophisticated second-stage separation of cast and wrought aluminium. This capability allows recyclers to create high-value, furnace-ready fractions that meet the stringent requirements of the secondary aluminium industry. This density-based sorting provides the vital foundation for purity, which is then further refined by complementary technologies to resolve even the most challenging material streams.

Our introduction of dynamic laser induced breakdown spectroscopy (LIBS) in 2023 significantly advanced the process by enabling high speed elemental analysis. By using a high precision laser to identify specific alloying elements, the Autosort Pulse delivers the sophisticated alloy separation required for premium secondary production. However, the chemical data from the laser can be enhanced even more when paired with visual AI that can identify the physical characteristics of the scrap.

Deep learning for high purity recovery

This is where Gainnext, our deep learning-based AI solution, becomes essential. By integrating this technology, recyclers can now distinguish visual characteristics such as shapes, sizes and dimensions that traditional sensors might miss. One of the earliest metals applications of Gainnext was upgrading wrought aluminium fractions. After the X-Tract removes high-alloy cast and high-density wrought aluminium, the resulting fraction typically still contains minor amounts of low-alloy cast. Gainnext provides the final refinement by processing thousands of images per millisecond to classify and remove the remaining low-alloy cast, along with other low-density contaminants such as UBCs. The result is a pure wrought fraction that commands higher market prices.

Solving the overlapping challenge with object singulation

AI also resolves the critical issue of overlapping material. High throughput often comes at the expense of sorting accuracy due to material overlap. When objects are too close to each other on the conveyor belt, traditional sensors can struggle to distinguish between individual objects. This usually results in a compromise: either reducing the throughput rate to ensure separation or accepting lower purity levels.

AI-based object singulation technology, integrated into the Autosort Pulse, resolves this by using neural networks to determine whether multiple pieces form a single object or several distinct ones. This innovation improves single object detection, allowing for higher throughput and faster processing speeds without sacrificing the quality of the final product.

The new FINDER: A modular platform for complex scrap

While software provides the logic, the hardware must be robust enough for the harsh environment of a scrap yard. Our newly launched Finder is a completely re-engineered metals sorting platform that incorporates built-in AI features to maximise recovery rates from even the most complex scrap streams. It addresses the need for high purity recovery through a modular, multi-sensor architecture that can be configured according to operational needs and further enhanced by Gainnext to resolve the most challenging material streams.

The foundation of the system is a powerful electromagnetic (EM) sensor for metal versus non-metal sorting. The real value lies in its unmatched flexibility; operators can integrate near infrared (NIR) sensors for high precision separation of printed circuit boards (PCBs) or the removal of insulated wire from the stream. This modularity is particularly effective for cleaning mixed aluminium fractions like Zorba or Twitch. Additionally, the optional Deep Laiser and RGB camera add object recognition capabilities that allow for the precise removal of non-metals, such as black plastics or coloured metals from mixed metal fractions. By removing these complex contaminants first, the Finder creates a cleaner feed that allows downstream X-Tract and Autosort Pulse units to work at peak efficiency.