MATERIALS / RECYCLING PETplanet Insider Vol. 25 No. 04/24 www.petpla.net 26 AI-powered object recognition for sorting food-grade from non-food-grade plastics Differentiation in detail Until now, food-grade sorting has proved a challenge for the industry as food and non-food packaging are often made of the same material and visually very similar which makes it difficult for sorting systems to differentiate and separate. Thanks to Tomra’s continued investment in a deep learning-based sorting add-on for its Autosort units, it is now possible for the first time to quickly and efficiently separate foodgrade from non-food-grade plastics for PET, PP and HDPE on a large scale. Tomra’s Gain technology, now rebranded Gainnext, resolves these challenges by further enhancing the sorting performance of the company’s Autosort units so they are capable of identifying objects that are hard and, in some cases, even impossible to classify using traditional optical waste sensors. Purity levels of over 95% By combining its traditional nearinfrared, visual spectrometry or other sensors with deep learning technology, Tomra has developed an accurate solution which achieves degrees of purity of 95% for the packaging applications in customers’ plants across UK and Europe. Tomra is also launching two nonfood applications that complement the company’s existing Gainnext ecosystem: an application for de-inking paper for cleaner paper streams, and a PET cleaner application for even higher purity PET bottle streams. Bottle-to-bottle quality Dr Volker Rehrmann, EVP, Head of Tomra Recycling, comments: “We have used AI technology to improve sorting performance for decades, but this latest groundbreaking application marks another industry first for us. AI has the power to transform resource recovery as we know it, and our latest sophisticated applications of deep learning and AI reinforce our position as a pioneer in this field. With its sophisticated use of deep learning, Gainnext enables food-grade sorting and bottle-to-bottle quality, tasks that have posed significant challenges for our industry for many years. The use of AI is driving material circularity at a time when it is needed most, with tightening regulations and increasing customer demand for technologically advanced solutions.” Indrajeed Prasad, Product Manager Deep Learning at Tomra Recycling, adds: “The use of deep learning technology not only automates manual sorting but also enables the industry to achieve high-quality recyclates through more granular sorting. Thanks to its ability to detect thousands of objects by material and shape in milliseconds, Gainnext solves even the most complex sorting tasks. Plus, with its integrated deep learning software, it offers the opportunity to adapt to future demands. We are delighted to be able to launch these innovative and muchneeded solutions to meet the ever more stringent quality requirements for sorting outputs, driven by the increasing demand from consumer brands for more high purity recycled content.” Field-proven technology Gainnext’s deep learning technology has been proven in the field for many years. Tomra introduced deep learning technology in 2019 with an application to identify and remove PEsilicon cartridges from PE streams. An application for wood chip classification soon followed in 2022. To date, more than 100 Autosort units with Gainnext are installed at material recovery facilities across the globe. Among the early adopters of the brand new applications are market-leading plants such as Berry Circular Polymers’ flagship facility in Leamington Spa, Viridor Avonmouth in Bristol, and the French Nord Pal Plast plant, which is owned by the European Dentis Group. www.tomra.com Leading global sorting solutions provider, Tomra Recycling, has announced the launch of three new applications to separate food-grade from non-food-grade plastics for PET, PP and HDPE. Following an invitation to the company’s site in Mülheim-Kärlich, Germany, PETplanet learned that the breakthrough was made possible by the company’s intensive research and development in deep learning, a subset of AI.
RkJQdWJsaXNoZXIy NTY0MjI=